Autonomous driving relies on robust models trained on large-scale, high-quality multi-view driving videos. Although world models provide a cost-effective solution for generating realistic driving data, they often suffer from identity drift, where the same object changes its appearance or category across frames due to the absence of instance-level temporal constraints. We introduce ConsisDrive, an identity-preserving driving world model designed to enforce temporal consistency at the instance level. Our framework incorporates two key components: (1) Instance-Masked Attention, which applies instance identity masks and trajectory masks within attention blocks to ensure that visual tokens interact only with their corresponding instance features across spatial and temporal dimensions, thereby preserving object identity consistency; and (2) Instance-Masked Loss, which adaptively emphasizes foreground regions with probabilistic instance masking, reducing background noise while maintaining overall scene fidelity. By integrating these mechanisms, ConsisDrive achieves state-of-the-art driving video generation quality and demonstrates significant improvements in downstream autonomous driving tasks on the nuScenes dataset.

We describe Generative Blocks World to interact with the scene of a generated image by manipulating simple geometric abstractions. Our method represents scenes as assemblies of convex 3D primitives, and the same scene can be represented by different numbers of primitives, allowing an editor to move either whole structures or small details. Once the scene geometry has been edited, the image is generated by a flow-based method, which is conditioned on depth and a texture hint. Our texture hint takes into account the modified 3D primitives, exceeding the texture-consistency provided by existing techniques. These texture hints (a) allow accurate object and camera moves and (b) preserve the identity of objects. Our experiments demonstrate that our approach outperforms prior works in visual fidelity, editability, and compositional generalization.

We develop a cost-efficient neurosymbolic agent to address challenging multi-turn image editing tasks such as "Detect the bench in the image while recoloring it to pink. Also, remove the cat for a clearer view and recolor the wall to yellow." It combines the fast, high-level subtask planning by large language models (LLMs) with the slow, accurate, tool-use, and local A* search per subtask to find a cost-efficient toolpath---a sequence of calls to AI tools. To save the cost of A* on similar subtasks, we perform inductive reasoning on previously successful toolpaths via LLMs to continuously extract/refine frequently used subroutines and reuse them as new tools for future tasks in an adaptive fast-slow planning, where the higher-level subroutines are explored first, and only when they fail, the low-level A* search is activated. The reusable symbolic subroutines considerably save exploration cost on the same types of subtasks applied to similar images, yielding a human-like fast-slow toolpath agent ``FaSTA*'': fast subtask planning followed by rule-based subroutine selection per subtask is attempted by LLMs at first, which is expected to cover most tasks, while slow A* search is only triggered for novel and challenging subtasks. By comparing with recent image editing approaches, we demonstrate FaSTA* is significantly more computationally efficient while remaining competitive with the state-of-the-art baseline in terms of success rate.

This paper introduces JavisDiT, a novel Joint Audio-Video Diffusion Trans- former designed for synchronized audio-video generation (JAVG). Based on the powerful Diffusion Transformer (DiT) architecture, JavisDiT simultaneously generates high-quality audio and video content from open-ended user prompts in a unified framework. To ensure audio-video synchronization, we introduce a fine-grained spatio-temporal alignment mechanism through a Hierarchical Spatial-Temporal Synchronized Prior (HiST-Sypo) Estimator. This module extracts both global and fine-grained spatio-temporal priors, guiding the synchronization between the visual and auditory components. Furthermore, we propose a new benchmark, JavisBench, which consists of 10,140 high-quality text-captioned sounding videos and focuses on synchronization evaluation in diverse and complex real-world scenarios. Further, we specifically devise a robust metric for measuring the synchrony between generated audio-video pairs in real-world content. Experimental results demonstrate that JavisDiT significantly outperforms existing methods by ensuring both high-quality generation and precise synchronization, setting a new standard for JAVG tasks. Our code, model, and data are available at https://javisverse.github.io/JavisDiT-page/.

Recent advances in diffusion-based video restoration (VR) demonstrate significant improvement in visual quality, yet yield a prohibitive computational cost during inference. While several distillation-based approaches have exhibited the potential of one-step image restoration, extending existing approaches to VR remains challenging and underexplored, particularly when dealing with high-resolution video in real-world settings. In this work, we propose a one-step diffusion-based VR model, termed as SeedVR2, which performs adversarial VR training against real data. To handle the challenging high-resolution VR within a single step, we introduce several enhancements to both model architecture and training procedures. Specifically, an adaptive window attention mechanism is proposed, where the window size is dynamically adjusted to fit the output resolutions, avoiding window inconsistency observed under high-resolution VR using window attention with a predefined window size. To stabilize and improve the adversarial post-training towards VR, we further verify the effectiveness of a series of losses, including a proposed feature matching loss without significantly sacrificing training efficiency. Extensive experiments show that SeedVR2 can achieve comparable or even better performance compared with existing VR approaches in a single step.

Video Foundation Models (VFMs) exhibit remarkable visual generation performance, but struggle in compositional scenarios (\eg, motion, numeracy, and spatial relation). In this work, we introduce Test-Time Optimization and Memorization (TTOM), a training-free framework that aligns VFM outputs with spatiotemporal layouts during inference for better text-image alignment. Rather than direct intervention to latents or attention per-sample in existing work, we integrate and optimize new parameters guided by a general layout-attention objective. Furthermore, we formulate video generation within a streaming setting, and maintain historical optimization contexts with a parametric memory mechanism that supports flexible operations, such as insert, read, update, and delete. Notably, we found that TTOM disentangles compositional world knowledge, showing powerful transferability and generalization. Experimental results on the T2V-CompBench and Vbench benchmarks establish TTOM as an effective, practical, scalable, and efficient framework to achieve cross-modal alignment for compositional video generation on the fly.

Flow-matching models deliver state-of-the-art fidelity in image and video generation, but the inherent sequential denoising process renders them slower. Existing acceleration methods like distillation, trajectory truncation, and consistency approaches are static, require retraining, and often fail to generalize across tasks. We propose FastFlow, a plug-and-play adaptive inference framework that accelerates generation in flow matching models. FastFlow identifies denoising steps that produce only minor adjustments to the denoising path and approximates them without using the full neural network models used for velocity predictions. The approximation utilizes finite-difference velocity estimates from prior predictions to efficiently extrapolate future states, enabling faster advancements along the denoising path at zero compute cost. This enables skipping computation at intermediary steps. We model the decision of how many steps to safely skip before requiring a full model computation as a multi-armed bandit problem. The bandit learns the optimal skips to balance speed with performance. FastFlow integrates seamlessly with existing pipelines and generalizes across image generation, video generation, and editing tasks. Experiments demonstrate a speedup of over $2.6\times$ while maintaining high-quality outputs. The source code for this work can be found at https://github.com/Div290/FastFlow.

Artificial Intelligence Generated Content (AIGC), particularly video generation with diffusion models, has been advanced rapidly. Invisible watermarking is a key technology for protecting AI-generated videos and tracing harmful content, and thus plays a crucial role in AI safety. Beyond post-processing watermarks which inevitably degrade video quality, recent studies have proposed distortion-free in-generation watermarking for video diffusion models. However, existing in-generation approaches are non-blind: they require maintaining all the message-key pairs and performing template-based matching during extraction, which incurs prohibitive computational costs at scale. Moreover, when applied to modern video diffusion models with causal 3D Variational Autoencoders (VAEs), their robustness against temporal disturbance becomes extremely weak. To overcome these challenges, we propose SIGMark, a Scalable In-Generation watermarking framework with blind extraction for video diffusion. To achieve blind-extraction, we propose to generate watermarked initial noise using a Global set of Frame-wise PseudoRandom Coding keys (GF-PRC), reducing the cost of storing large-scale information while preserving noise distribution and diversity for distortion-free watermarking. To enhance robustness, we further design a Segment Group-Ordering module (SGO) tailored to causal 3D VAEs, ensuring robust watermark inversion during extraction under temporal disturbance. Comprehensive experiments on modern diffusion models show that SIGMark achieves very high bit-accuracy during extraction under both temporal and spatial disturbances with minimal overhead, demonstrating its scalability and robustness. Our code is available at https://github.com/JeremyZhao1998/SIGMark-release.

The rapid development of visual generative models raises the need for more scalable and human-aligned evaluation methods. While the crowdsourced Arena platforms offer human preference assessments by collecting human votes, they are costly and time-consuming, inherently limiting their scalability. Leveraging vision-language model (VLMs) as substitutes for manual judgments presents a promising solution. However, the inherent hallucinations and biases of VLMs hinder alignment with human preferences, thus compromising evaluation reliability. Additionally, the static evaluation approach lead to low efficiency. In this paper, we propose K-Sort Eval, a reliable and efficient VLM-based evaluation framework that integrates posterior correction and dynamic matching. Specifically, we curate a high-quality dataset from thousands of human votes in K-Sort Arena, with each instance containing the outputs and rankings of K models. When evaluating a new model, it undergoes (K+1)-wise free-for-all comparisons with existing models, and the VLM provide the rankings. To enhance alignment and reliability, we propose a posterior correction method, which adaptively corrects the posterior probability in Bayesian updating based on the consistency between the VLM prediction and human supervision. Moreover, we propose a dynamic matching strategy, which balances uncertainty and diversity to maximize the expected benefit of each comparison, thus ensuring more efficient evaluation. Extensive experiments show that K-Sort Eval delivers evaluation results consistent with K-Sort Arena, typically requiring fewer than 90 model runs, demonstrating both its efficiency and reliability. The dataset and code are publicly available.

Unified multimodal models (UMMs) unify visual understanding and generation within a single architecture. However, conventional training relies on image–text pairs (or sequences) whose captions are typically sparse and miss fine-grained visual details, even when they use hundreds of words to describe a simple image. We introduce Reconstruction Alignment (RecA), a resource-efficient post-training method that leverages visual understanding encoder embeddings as dense “text prompts,” providing rich supervision without captions. Concretely, RecA conditions a UMM on its own visual understanding embeddings and optimizes it to reconstruct the input image with a self-supervised reconstruction loss, thereby realigning understanding and generation. Despite its simplicity, RecA is broadly applicable: across autoregressive, masked-autoregressive, and diffusion-based UMMs, it consistently improves generation and editing fidelity. With only 27 GPU-hours, post-training with RecA substantially improves image generation performance on GenEval (0.73 → 0.90) and DPGBench (80.93 → 88.15), while also boosting editing benchmarks (ImgEdit 3.38 → 3.75, GEdit 6.94 → 7.27). Notably, RecA surpasses much larger open-source models and applies broadly across diverse UMM architectures, establishing it as an efficient and general post-training alignment strategy for UMMs.

Existing AI-based film generation systems can generate high-quality videos, but struggle to design expressive camera language and establish cinematic rhythm. This deficiency leads to templated visuals and unengaging narratives. To address these limitations, we introduce FilMaster, an end-to-end automated film generation system that integrates real-world cinematic principles to generate professional-grade, editable films. Inspired by professional filmmaking, FilMaster is built on two key cinematic principles: (1) camera language design by learning cinematography from extensive real-world film references, and (2) cinematic rhythm by emulating professional post-production workflows. For camera language, our Multi-shot Synergized Camera Language Design module introduces a novel scene-level Retrieval-Augmented Generation (RAG) framework. Unlike shot-level RAG which retrieves references independently and often leads to visual incoherence, our approach treats an entire scene, comprising multiple shots with a shared spatio-temporal context and narrative objective, as a single, unified query. This holistic query retrieves a consistent set of semantically similar shots with cinematic techniques from a large corpus of 440,000 real film clips. These references then guide an LLM to synergistically plan coherent and expressive camera language for all shots within that scene. To achieve cinematic rhythm, our Audience-Aware Cinematic Rhythm Control module emulates professional post-production, featuring a Rough Cut assembly followed by a Fine Cut process that uses simulated audience feedback to optimize the integration of video and sound for cinematic rhythm. Extensive experiments show superior performance in camera language and cinematic rhythm, paving the way for generative AI in professional filmmaking.

Recent AIGC advances have rapidly expanded from text-to-image generation toward high-quality multimodal synthesis across video and audio. Within this context, joint audio-video generation (JAVG) has emerged as a fundamental task that produces synchronized and semantically aligned sound and vision from textual descriptions. However, compared with advanced commercial models such as Veo3, existing open-source methods still suffer from limitations in generation quality, temporal synchrony, and alignment with human preferences. To bridge the gap, this paper presents JavisDiT++, a concise yet powerful framework for efficient and effective JAVG. First, we introduce a modality-specific mixture-of-experts (MS-MoE) design that enables cross-modal interaction efficacy while enhancing single-modal generation quality. Then, we propose a temporal-aligned RoPE (TA-RoPE) strategy to achieve explicit, frame-level synchronization between audio and video tokens. Besides, we develop an audio-video direct preference optimization (AV-DPO) method to align model outputs with human preference across quality, consistency, and synchrony dimensions. Built upon Wan2.1-1.3B-T2V, our model achieves state-of-the-art performance merely with around 1M public training entries, significantly outperforming prior approaches in both qualitative and quantitative evaluations. Comprehensive ablation studies have been conducted to validate the effectiveness of our proposed modules. All the code, model, and dataset are released at https://JavisVerse.github.io/JavisDiT2-page.

High-precision controllable remote sensing image generation is both meaningful and challenging. Existing diffusion models often produce low-fidelity objects due to their inability to adequately capture morphological details, which may affect the robustness and reliability of object detection models. To enhance the accuracy and fidelity of generated objects in remote sensing, this paper proposes Object Fidelity Diffusion (OF-Diff), which effectively improves the fidelity of generated objects. Specifically, we are the first to extract the prior shapes of objects based on the layout for diffusion models in remote sensing. Then, we introduce a self-distillation diffusion model with consistency distillation loss, which can generate high-fidelity remote sensing images without providing real images during the sampling phase. Furthermore, we introduce DDPO to fine-tune the diffusion process, making the generated remote sensing images more diverse and semantically consistent. Comprehensive experiments demonstrate that OF-Diff outperforms state-of-the-art methods in the remote sensing across key quality metrics. Notably, the performance of several polymorphic and small object classes shows significant improvement. For instance, the mAP increases by 8.3%, 7.7%, and 4.0% for airplanes, ships, and vehicles, respectively.

We introduce SANA-Video, a small diffusion model that can efficiently generate videos up to 720×1280 resolution and minute-length duration. SANA-Video synthesizes high-resolution, high-quality and long videos with strong text-video alignment at a remarkably fast speed, deployable on RTX 5090 GPU. Two core designs ensure our efficient, effective and long video generation: (1) Linear DiT: We leverage linear attention as the core operation, which is more efficient than vanilla attention given the large number of tokens processed in video generation. (2) Constant-Memory KV cache for Block Linear Attention: we design block-wise autoregressive approach for long video generation by employing a constant-memory state, derived from the cumulative properties of linear attention. This KV cache provides the Linear DiT with global context at a fixed memory cost, eliminating the need for a traditional KV cache and enabling efficient, minute-long video generation. In addition, we explore effective data filters and model training strategies, narrowing the training cost to 12 days on 64 H100 GPUs, which is only 1\% of the cost of MovieGen. Given its low cost, SANA-Video achieves competitive performance compared to modern state-of-the-art small diffusion models (e.g., Wan 2.1-1.3B and SkyReel-V2-1.3B) while being 16x faster in measured latency. Moreover, SANA-Video can be deployed on RTX 5090 GPUs with NVFP4 precision, accelerating the inference speed of generating a 5-second 720p video from 71s to 29s (2.4x} speedup). In summary, SANA-Video enables low-cost, high-quality video generation. Code and model will be publicly released.

While thinking-aware generation aims to improve performance on complex tasks, we identify a critical failure mode where existing sequential, autoregressive approaches can paradoxically degrade performance due to error propagation. To systematically analyze this issue, we propose ParaBench, a new benchmark designed to evaluate both text and image output modalities. Our analysis using ParaBench reveals that this performance degradation is strongly correlated with poor alignment between the generated reasoning and the final image. To resolve this, we propose a parallel multimodal diffusion framework, MMaDA-Parallel, that enables continuous, bidirectional interaction between text and images throughout the entire denoising trajectory. MMaDA-Parallel is trained with supervised finetuning and then further optimized by Parallel Reinforcement Learning (ParaRL), a novel strategy that applies semantic rewards along the trajectory to enforce cross-modal consistency. Experiments validate that our model significantly improves cross-modal alignment and semantic consistency, achieving a 6.9\% improvement in Output Alignment on ParaBench compared to the state-of-the-art model, Bagel, establishing a more robust paradigm for thinking-aware image synthesis.

We propose a novel approach to image generation by decomposing an image into a structured sequence, where each element in the sequence shares the same spatial resolution but differs in the number of unique tokens used, capturing different level of visual granularity. Image generation is carried out through our newly introduced Next Visual Granularity (NVG) generation framework, which generates a visual granularity sequence beginning from an empty image and progressively refines it, from global layout to fine details, in a structured manner. This iterative process encodes a hierarchical, layered representation that offers fine-grained control over the generation process across multiple granularity levels. We train a series of NVG models for class-conditional image generation on the ImageNet dataset and observe clear scaling behavior. Compared to the VAR series, NVG consistently outperforms it in terms of FID scores (3.30 $\rightarrow$ 3.03, 2.57 $\rightarrow$ 2.44, 2.09 $\rightarrow$ 2.06). We also conduct extensive analysis to showcase the capability and potential of the NVG framework. Our code and models are released at https://yikai-wang.github.io/nvg.

Unified multimodal models (UMMs) have emerged as a powerful paradigm for seamlessly unifying text and image understanding and generation. However, prevailing evaluations treat these abilities in isolation, such that tasks with multimodal inputs and outputs are scored primarily through unimodal reasoning, i.e., textual benchmarks emphasize language-based reasoning, while visual benchmarks emphasize reasoning outcomes manifested in the pixels. We introduce ROVER to address this pressing need to test reciprocal cross-modal reasoning, the use of one modality to guide, verify, or refine outputs in the other, an ability central to the vision of unified multimodal intelligence. ROVER is a human-annotated benchmark that explicitly targets reciprocal cross-modal reasoning, which contains 1,312 tasks grounded in 1,876 images, spanning two complementary settings. Verbally-augmented reasoning for visual generation evaluates whether models can use verbal prompts and reasoning chains to guide faithful image synthesis. Visually-augmented reasoning for verbal generation evaluates whether models can generate intermediate visualizations that strengthen their own reasoning processes for question answering. Experiments on 17 unified models reveal two key findings: (i) Cross-modal reasoning determines visual generation quality, with interleaved models significantly outperforming non-interleaved ones; notably, combining strong unimodal models fails to achieve comparable reasoning. (ii) Models show dissociation between physical and symbolic reasoning: they succeed at interpreting perceptual concepts literally but fail to construct visual abstractions for symbolic tasks, where faulty reasoning harms performance. These results highlight reciprocal cross-modal reasoning as a critical frontier for enabling true omnimodal generation. Homepage: https://roverbench.github.io

Recent years have witnessed remarkable progress in generative AI, with natural language emerging as the most common conditioning input. As underlying models grow more powerful, researchers are exploring increasingly diverse conditioning signals -- such as depth maps, edge maps, camera parameters, and reference images -- to give users finer control over generation. Among different modalities, sketches constitute a natural and long-standing form of human communication, enabling rapid expression of visual concepts. Yet algorithms that effectively handle true freehand sketches -- with their inherent abstraction and distortions -- remain largely unexplored. In this work, we distinguish between edge maps, often regarded as “sketches” in the literature, and genuine freehand sketches. We pursue the challenging goal of balancing photorealism with sketch adherence when generating images from freehand input. A key obstacle is the absence of ground-truth, pixel-aligned images: by their nature, freehand sketches do not have a single correct alignment. To address this, we propose a modulation-based approach that prioritizes semantic interpretation of the sketch over strict adherence to individual edge positions. We further introduce a novel loss that enables training on freehand sketches without requiring ground-truth pixel-aligned images. We show that our method outperforms existing approaches in both semantic alignment with freehand sketch inputs and in the realism and overall quality of the generated images.

In this paper, we propose a 3D asset-referenced diffusion model for image generation, exploring how to integrate 3D assets into image diffusion models. Existing reference-based image generation methods leverage large-scale pretrained diffusion models and demonstrate strong capability in generating diverse images conditioned on a single reference image. However, these methods are limited to single-image references and cannot leverage 3D assets, constraining their practical versatility. To address this gap, we present a cross-domain diffusion model with dual-branch perception that leverages multi-view RGB images and point maps of 3D assets to jointly model their colors and canonical-space coordinates, achieving precise consistency between generated images and the 3D references. Our spatially aligned dual-branch generation architecture and domain-decoupled generation mechanism ensure the simultaneous generation of two spatially aligned but content-disentangled outputs, RGB images and point maps, linking 2D image attributes with 3D asset attributes. Experiments show that our approach effectively uses 3D assets as references to produce images consistent with the given assets, opening new possibilities for combining diffusion models with 3D content creation.

Realistic talking-head video generation is critical for virtual avatars, film production, and interactive systems. Current methods struggle with nuanced emotional expressions due to the lack of fine-grained emotion control. To address this issue, we introduce a novel two-stage method (AUHead) to disentangle fine-grained emotion control, i.e. , Action Units (AUs), from audio and achieve controllable generation. In the first stage, we explore the AU generation abilities of large audio-language models (ALMs), by spatial-temporal AU tokenization and an "emotion-then-AU" chain-of-thought mechanism. It aims to disentangle AUs from raw speech, effectively capturing subtle emotional cues. In the second stage, we propose an AU-driven controllable diffusion model that synthesizes realistic talking-head videos conditioned on AU sequences. Specifically, we first map the AU sequences into the structured 2D facial representation to enhance spatial fidelity, and then model the AU-vision interaction within cross-attention modules. To achieve flexible AU-quality trade-off control, we introduce an AU disentanglement guidance strategy during inference, further refining the emotional expressiveness and identity consistency of the generated videos. Results on benchmark datasets demonstrate that our approach achieves competitive performance in emotional realism, accurate lip synchronization, and visual coherence, significantly surpassing existing techniques. Our implementation is available at https://github.com/laura990501/AUHead_ICLR

Diffusion models have shown impressive performance in many visual generation and manipulation tasks. Many existing methods focus on training a model for a specific task, especially, text-to-video (T2V) generation, while many other works focus on finetuning the pretrained T2V model for image-to-video (I2V), video-to-video (V2V), image and video manipulation tasks, \etc. However, training a strong T2V foundation model requires a large amount of high-quality annotations, which is very costly. In addition, many existing models can perform only one or several tasks. In this work, we introduce a unified framework, namely \textit{many-for-many}, which leverages the available training data from many different visual generation and manipulation tasks to train a single model for those different tasks. Specifically, we design a lightweight adapter to unify the different conditions in different tasks, then employ a joint image-video learning strategy to progressively train the model from scratch. Our joint learning not only leads to a unified generation and manipulation model but also benefits the performance of different tasks. In addition, we introduce depth maps as a condition to help our model better perceive the 3D space in visual generation. Two versions of our model are trained with different model sizes (8B and 2B), each of which can perform more than 10 different tasks. In particular, our 8B model demonstrates highly competitive performance in different generation and manipulation tasks compared to open-source and even commercial engines. Our models and source codes will be made publicly available.

Recent Text-to-Video (T2V) models have demonstrated powerful capability in visual simulation of real-world geometry and physical laws, indicating its potential as implicit world models. Inspired by this, we explore the feasibility of leveraging the video generation prior for viewpoint planning from given 4D scenes, since videos internally accompany dynamic scenes with natural viewpoints. To this end, we propose a two-stage paradigm to adapt pre-trained T2V models for viewpoint prediction, in a compatible manner. First, we inject the 4D scene representation into the pre-trained T2V model via an adaptive learning branch, where the 4D scene is viewpoint-agnostic and the conditional generated video embeds the viewpoints visually. Then, we formulate viewpoint extraction as a hybrid-condition guided camera extrinsic denoising process. Specifically, a camera extrinsic diffusion branch is further introduced onto the pre-trained T2V model, by taking the generated video and 4D scene as input. Experimental results show the superiority of our proposed method over existing competitors, and ablation studies validate the effectiveness of our key technical designs. To some extent, this work proves the potential of video generation models toward 4D interaction in real world.

In-context image editing aims to modify images based on a contextual sequence comprising text and previously generated images. Existing methods typically depend on task-specific pipelines and expert models (e.g., segmentation and inpainting) to curate training data. In this work, we explore whether an in-context image editing model can be learned directly from videos. We introduce a scalable approach to annotate videos as interleaved multimodal sequences. To effectively learn from this data, we design three proxy tasks: next-image prediction, current segmentation prediction, and next-segmentation prediction. Additionally, we propose a novel multi-turn image editing benchmark to advance research in this area. Extensive experiments demonstrate that our model exhibits strong in-context image editing capabilities and achieves state-of-the-art results on two multi-turn image editing benchmarks. Despite being trained exclusively on videos, our model also shows promising abilities in multi-concept composition, story generation, and chain-of-editing applications.

Drag-based image editing has long suffered from distortions in the target region, largely because the priors of earlier base models, Stable Diffusion, are insufficient to project optimized latents back onto the natural image manifold. With the shift from UNet-based DDPMs to more scalable DiT with flow matching (e.g., SD3.5, FLUX), generative priors have become significantly stronger, enabling advances across diverse editing tasks. However, drag-based editing has yet to benefit from these stronger priors. This work introduces DragFlow, the first framework to effectively harness FLUX’s rich prior via region-based supervision, enabling full use of its finer-grained, spatially precise features for drag-based editing and achieving substantial improvements over existing baselines. We first show that directly applying point-based drag editing to DiTs performs poorly: unlike the highly compressed features of UNets, DiT features are insufficiently structured to provide reliable guidance for point-wise motion supervision. To overcome this limitation, DragFlow introduces a region-based editing paradigm, where affine transformations enable richer and more consistent feature supervision. Additionally, we integrate pretrained open-domain personalization adapters (e.g., IP-Adapter) to enhance subject consistency, while preserving background fidelity through gradient mask-based hard constraints. Multimodal large language models (MLLMs) are further employed to resolve task ambiguities. For evaluation, we curate a novel Region-based Dragging benchmark (ReD Bench) featuring region-level dragging instructions. Extensive experiments on DragBench-DR and ReD Bench show that DragFlow surpasses both point-based and region-based baselines, setting a new state-of-the-art in drag-based image editing. Code and dataset are available at https://github.com/Edennnnnnnnnn/DragFlow.

Diffusion models have achieved impressive results in generating high-quality images. Yet, they often struggle to faithfully align the generated images with the input prompts. This limitation is associated with synchronous denoising, where all pixels simultaneously evolve from random noise to clear images. As a result, during generation, the prompt-related regions can only reference the unrelated regions at the same noise level, failing to obtain clear context and ultimately impairing text-to-image alignment. To address this issue, we propose asynchronous diffusion models, a novel framework that allocates distinct timesteps to different pixels and reformulates the pixel-wise denoising process. By dynamically modulating the timestep schedules of individual pixels, prompt-related regions are denoised more gradually than unrelated regions, thereby allowing them to leverage clearer inter-pixel context. Consequently, these prompt-related regions achieve better alignment in the final images. Extensive experiments demonstrate that our asynchronous diffusion models can significantly improve text-to-image alignment across diverse prompts. The code repository for this work is available at https://github.com/hu-zijing/AsynDM.

Large pretrained diffusion models have significantly enhanced the quality of generated videos, and yet their use in real-time streaming remains limited. Autoregressive models offer a natural framework for sequential frame synthesis but require heavy computation to achieve high fidelity. Diffusion distillation can compress these models into efficient few-step variants, but existing video distillation approaches largely adapt image-specific methods that neglect temporal dependencies. These techniques often excel in image generation but underperform in video synthesis, exhibiting reduced motion coherence, error accumulation over long sequences, and a latency-quality trade-off. We identify two factors that result in these limitations: insufficient utilization of temporal context during step reduction and implicit prediction of subsequent noise levels in next-chunk prediction (i.e., exposure bias). To address these issues, we propose Diagonal Distillation, which operates orthogonally to existing approaches and better exploits temporal information across both video chunks and denoising steps. Central to our approach is an asymmetric generation strategy: more steps early, fewer steps later. This design allows later chunks to inherit rich appearance information from thoroughly processed early chunks, while using partially denoised chunks as conditional inputs for subsequent synthesis. By aligning the implicit prediction of subsequent noise levels during chunk generation with the actual inference conditions, our approach mitigates error propagation and reduces oversaturation in long-range sequences. We further incorporate implicit optical flow modeling to preserve motion quality under strict step constraints. Our method generates a 5-second video in 2.61 seconds (up to 31 FPS), achieving a 277.3× speedup over the undistilled model.

Current image immunization defense techniques against diffusion-based editing embed imperceptible noise into target images to disrupt editing models. However, these methods face scalability challenges, as they require time-consuming optimization for each image separately, taking hours for small batches. To address these challenges, we introduce DiffVax, a scalable, lightweight, and optimization-free framework for image immunization, specifically designed to prevent diffusion-based editing. Our approach enables effective generalization to unseen content, reducing computational costs and cutting immunization time from days to milliseconds, achieving a speedup of 250,000x. This is achieved through a loss term that ensures the failure of editing attempts and the imperceptibility of the perturbations. Extensive qualitative and quantitative results demonstrate that our model is scalable, optimization-free, adaptable to various diffusion-based editing tools, robust against counter-attacks, and, for the first time, effectively protects video content from editing. More details are available in https://diffvax.github.io/.

Recent hybrid video generation models combine autoregressive temporal dynamics with diffusion-based spatial denoising, but their sequential, iterative nature leads to error accumulation and long inference times. In this work, we propose a distillation-based framework for efficient causal video generation that enables high-quality synthesis with extreme limited denoising steps. Our approach builds upon Distribution Matching Distillation (DMD) framework and proposes a novel form of Adversarial Self-Distillation (ASD) strategy, which aligns the outputs of the student model's $n$-step denoising process with its $(n+1)$-step version in the distribution level. This design provides smoother supervision by bridging small intra-student gaps and more informative guidance by combining teacher knowledge with locally consistent student behavior, substantially improving training stability and generation quality in extremely few-step scenarios. In addition, we present a First-Frame Enhancement (FFE) strategy, which allocates more denoising steps to the initial frames to mitigate error propagation while applying larger skipping steps to later frames. Extensive experiments on VBench demonstrate that our method surpasses state-of-the-art approaches in both one-step and two-step video generation. Notably, our framework produces a single distilled model that flexibly supports multiple inference-step settings, eliminating the need for repeated re-distillation and enabling efficient, high-quality video synthesis.

Recent advances in video diffusion models shows promise for generating robotic decision-making data, with trajectory conditions further enabling fine-grained control. However, existing methods primarily focus on individual object motion and struggle to capture multi-object interaction crucial in complex manipulation. This limitation arises from entangled features in overlapping regions, leading to degraded visual fidelity. To address this, we present RoboMaster, a novel framework that models inter-object dynamics via a collaborative trajectory formulation. Unlike prior methods that decompose objects, our core is to decompose the interaction process into three sub-stages: pre-interaction, interaction, and post-interaction, and models each phase using the dominant object, specifically the robotic arm in the pre- and post-interaction phases and the manipulated object during interaction. This design effectively alleviates the multi-object feature fusion issue in prior work. To further ensure subject semantic consistency across the video, we incorporate appearance- and shape-aware latent representations for objects. Extensive experiments on the challenging Bridge dataset, as well as RLBench and SIMPLER benchmarks, demonstrate that our method establishs new state-of-the-art performance in trajectory-controlled video generation for robotic manipulation. Project Page: https://fuxiao0719.github.io/projects/robomaster/

Prevailing autoregressive (AR) models for text-to-image generation either rely on heavy, computationally-intensive diffusion models to process continuous image tokens, or employ vector quantization (VQ) to obtain discrete tokens with quantization loss. In this paper, we push the autoregressive paradigm forward with NextStep-1, a 14B autoregressive model paired with a 157M flow matching head, trained on discrete text tokens and continuous image tokens with next-token prediction objectives. NextStep-1 achieves state-of-the-art performance for autoregressive models in text-to-image generation tasks, exhibiting strong capabilities in high-fidelity image synthesis. Furthermore, our method shows strong performance in image editing, highlighting the power and versatility of our unified approach. To facilitate open research, we have released our code and models to the community at https://github.com/stepfun-ai/NextStep-1.

Most existing learning-based methods for solving imaging inverse problems can be roughly divided into two classes: iterative algorithms, such as plug-and-play and diffusion methods leveraging pretrained denoisers, and unrolled architectures that are trained end-to-end for specific imaging problems. Iterative methods in the first class are computationally costly and often yield suboptimal reconstruction performance, whereas unrolled architectures are generally problem-specific and require expensive training. In this work, we propose a novel non-iterative, lightweight architecture that incorporates knowledge about the forward operator (acquisition physics and noise parameters) without relying on unrolling. Our model is trained to solve a wide range of inverse problems, such as deblurring, magnetic resonance imaging, computed tomography, inpainting, and super-resolution, and handles arbitrary image sizes and channels, such as grayscale, complex, and color data. The proposed model can be easily adapted to unseen inverse problems or datasets with a few fine-tuning steps (up to a few images) in a self-supervised way, without ground-truth references. Throughout a series of experiments, we demonstrate state-of-the-art performance from medical imaging to low-photon imaging and microscopy. Our code is available at https://github.com/matthieutrs/ram.

Character image animation, which synthesizes videos of reference characters driven by pose sequences, has advanced rapidly but remains largely limited to single-human settings. Existing methods struggle to generalize to multi-humanoid scenarios, which involve diverse humanoid forms, complex interactions, and frequent occlusions. We address this gap with two key innovations. First, we introduce unified motion representations that extract identity-agnostic motions and explicitly bind them to corresponding characters, enabling generalization across diverse humanoid forms and seamless extension to multi-humanoid scenarios. Second, we propose a holistic 4D-anchored paradigm that constructs a shared 4D space to fuse motion representations with video latents, and further reinforces this process with hierarchical 4D-level supervision to better handle interactions and occlusions. We instantiate these ideas in MotionWeaver, an end-to-end framework for multi-humanoid image animation. To support this setting, we curate a 46-hour dataset of multi-human videos with rich interactions, and construct a 300-video benchmark featuring paired humanoid characters. Quantitative and qualitative experiments demonstrate that MotionWeaver not only achieves state-of-the-art results on our benchmark but also generalizes effectively across diverse humanoid forms, complex interactions, and challenging multi-humanoid scenarios.

Recent studies have explored autoregressive models for image generation, with promising results, and have combined diffusion models with autoregressive frameworks to optimize image generation via diffusion losses. In this study, we present a theoretical analysis of diffusion and autoregressive models with diffusion loss, highlighting the latter's advantages. We present a theoretical comparison of conditional diffusion and autoregressive diffusion with diffusion loss, demonstrating that patch denoising optimization in autoregressive models effectively mitigates condition errors and leads to a stable condition distribution. Our analysis also reveals that autoregressive condition generation refines the condition, causing the condition error influence to decay exponentially. In addition, we introduce a novel condition refinement approach based on Optimal Transport (OT) theory to address ``condition inconsistency''. We theoretically demonstrate that formulating condition refinement as a Wasserstein Gradient Flow ensures convergence toward the ideal condition distribution, effectively mitigating condition inconsistency. Experiments demonstrate the superiority of our method over diffusion and autoregressive models with diffusion loss methods.

We present MeanCache, a training-free caching framework for efficient Flow Matching inference. Existing caching methods reduce redundant computation but typically rely on instantaneous velocity information (e.g., feature caching), which often leads to severe trajectory deviations and error accumulation under high acceleration ratios. MeanCache introduces an average-velocity perspective: by leveraging cached Jacobian--vector products (JVP) to construct interval average velocities from instantaneous velocities, it effectively mitigates local error accumulation. To further improve cache timing and JVP reuse stability, we develop a trajectory-stability scheduling strategy as a practical tool, employing a Peak-Suppressed Shortest Path under budget constraints to determine the schedule. Experiments on FLUX.1, Qwen-Image, and HunyuanVideo demonstrate that MeanCache achieves $4.12\times$, $4.56\times$, and $3.59\times$ acceleration, respectively, while consistently outperforming state-of-the-art caching baselines in generation quality. We believe this simple yet effective approach provides a new perspective for Flow Matching inference and will inspire further exploration of stability-driven acceleration in commercial-scale generative models.

In recent years, integrating multimodal understanding and generation into a single unified model has emerged as a promising paradigm. While this approach achieves strong results in text-to-image (T2I) generation, it still struggles with precise image editing. We attribute this limitation to an imbalanced division of responsibilities. The understanding module primarily functions as a translator that encodes user instructions into semantic conditions, while the generation module must simultaneously act as designer and painter, inferring the original layout, identifying the target editing region, and rendering the new content. This imbalance is counterintuitive because the understanding module is typically trained with several times more data on complex reasoning tasks than the generation module. To address this issue, we introduce Draw-In-Mind (DIM), a dataset comprising two complementary subsets: (i) DIM-T2I, containing 14M long-context image-text pairs to enhance complex instruction comprehension; and (ii) DIM-Edit, consisting of 233K chain-of-thought imaginations generated by GPT-4o, serving as explicit design blueprints for image edits. We connect a frozen Qwen2.5-VL-3B with a trainable SANA1.5-1.6B via a lightweight two-layer MLP, and train it on the proposed DIM dataset, resulting in DIM-4.6B-T2I/Edit. Despite its modest parameter scale, DIM-4.6B-Edit achieves SOTA or competitive performance on the ImgEdit and GEdit-Bench benchmarks, outperforming much larger models such as UniWorld-V1 and Step1X-Edit. These findings demonstrate that explicitly assigning the design responsibility to the understanding module provides significant benefits for image editing. Our dataset and models are available at https://github.com/showlab/DIM.

Diffusion Transformer has shown remarkable abilities in generating high-fidelity videos, delivering visually coherent frames and rich details over extended durations. However, existing video generation models still fall short in subject-consistent video generation due to an inherent difficulty in parsing prompts that specify complex spatial relationships, temporal logic, and interactions among multiple subjects. To address this issue, we propose BindWeave, a unified framework that handles a broad range of subject-to-video scenarios from single-subject cases to complex multi-subject scenes with heterogeneous entities. To bind complex prompt semantics to concrete visual subjects, we introduce an MLLM-DiT framework in which a pretrained multimodal large language model performs deep cross-modal reasoning to ground entities and disentangle roles, attributes, and interactions, yielding subject-aware hidden states that condition the diffusion transformer for high-fidelity subject-consistent video generation. Experiments on the OpenS2V benchmark demonstrate that our method achieves superior performance across subject consistency, naturalness, and text relevance in generated videos, outperforming existing open-source and commercial models.

Handwritten Text Generation (HTG) seeks to synthesize realistic and personalized handwriting by modeling stylistic and structural traits. While recent diffusion-based approaches have advanced generation fidelity, they typically rely on auxiliary style or content encoders with handcrafted objectives, leading to complex training pipelines and limited interaction across factors. In this work, we present InkSpire, a diffusion transformer based model that unifies style, content, and noise within a shared latent space. By eliminating explicit encoders, InkSpire streamlines optimization while enabling richer feature interaction and stronger in-context generation. To further enhance flexibility, we introduce a multi-line masked infilling strategy that allows training directly on raw text-line images, together with a revised positional encoding that supports arbitrary-length multi-line synthesis and fine-grained character editing. Moreover, InkSpire is trained on a bilingual Chinese–English corpus, enabling a single model to handle both Chinese and English handwriting generation with high fidelity and stylistic diversity, thereby overcoming the need for language-specific systems. Extensive experiments on IAM and ICDAR2013 demonstrate that InkSpire achieves superior structural accuracy and stylistic diversity compared to prior state-of-the-art methods.

Camera control is important in text-to-video generation for achieving realistic scene navigation and view synthesis. This control is defined by parameters that describe movement through 3D space, thereby introducing 3D consistency into the generation process. A core challenge for existing methods is achieving 3D consistency within the 2D pixel domain. Strategies that directly integrate camera conditions into text-to-video models often produce artifacts, while those relying on explicit 3D supervision face challenges with generalization. Both limitations originate from the gap between the 2D pixel space and the underlying 3D world. The key insight is that the projection of a smooth 3D camera movement produces consistency in object view, appearance, and motion across 2D frames. Inspired by this insight, we propose MoCa, a dual-branch framework that bridges this gap by modeling object consistency to implicitly learn 3D relationships between the camera and the scene. To ensure view consistency, we design a Spatial-Temporal Camera Encoder with Plücker embedding, which encodes camera trajectories into a geometrically grounded latent representation. For appearance consistency, we introduce a semantic guidance strategy that leverages persistent vision-language features to maintain object identity and texture across frames. To address motion consistency, we propose an object-aware motion disentanglement mechanism that separates object dynamics from global camera movement, ensuring precise camera control and natural object motion. Experiments show that MoCa achieves accurate camera control while preserving video quality, offering a practical and effective solution for camera-controllable video generation.

We introduce *Generative Universal Verifier*, a novel concept and plugin designed for next-generation multimodal reasoning in vision-language models and unified multimodal models, providing the fundamental capability of reflection and refinement on visual outcomes during the reasoning and generation process. This work makes three main contributions: (1) We build **ViVerBench**, a comprehensive benchmark spanning $16$ categories of critical tasks for evaluating visual outcomes in multimodal reasoning. Results show that existing VLMs consistently underperform across these tasks, underscoring a substantial gap from human-level capability in reliable visual verification. (2) We design two automated pipelines to construct large-scale visual verification data and train **OmniVerifier-7B**, the first omni-capable generative verifier trained for universal visual verification and achieves notable gains on ViVerBench(+$8.3$). Through training, we identify three atomic capabilities in visual verification and demonstrate how they generalize and interact synergistically. (3) We propose **OmniVerifier-TTS**, a sequential test-time scaling paradigm that leverages the universal verifier to bridge image generation and editing within unified models, enhancing the upper bound of generative ability through iterative fine-grained optimization. Beyond generation, we extend universal verifier to broader world-modeling interleaved reasoning scenarios. Empirically, OmniVerifier-TTS achieves improvements on T2I-ReasonBench(+$3.7$), and GenEval++(+$4.3$), outperforming existing parallel test-time scaling methods, such as Best-of-N. By endowing multimodal reasoning with reliable visual verification, OmniVerifier advances both reliable reflection during generation and scalable test-time refinement, marking a step toward more trustworthy and controllable next-generation reasoning systems.

Image personalization enables customizing Text-to-Image models with a few reference images but is plagued by "concept coupling"—the model creating spurious associations between a subject and its context. Existing methods tackle this indirectly, forcing a trade-off between personalization fidelity and text control. This paper is the first to formalize concept coupling as a statistical dependency problem, identifying two root causes: a Denoising Dependence Discrepancy that arises during the generative process, and a Prior Dependence Discrepancy within the learned concept itself. To address this, we introduce ACCORD, a framework with two targeted, plug-and-play regularization losses. The Denoising Decouple Loss minimizes dependency changes across denoising steps, while the Prior Decouple Loss aligns the concept’s relational priors with those of its superclass. Extensive experiments across subject, style, and face personalization demonstrate that ACCORD achieves a superior balance between fidelity and text control, consistently improving upon existing methods.

Autoregressive (AR) models have emerged as a powerful framework for image generation, yet they remain bound by a fundamental limitation: once a prediction is made, it cannot be revised. Each step marches forward in a strict left-to-right sequence, causing small errors to accumulate and compromise the final image. In this work, we reimagine this process with TensorAR, a decoder-only AR model that shifts from predicting discrete tokens to predicting overlapping tensor windows. This simple change transforms image synthesis into a process of next-tensor prediction, enabling the model to refine earlier outputs while preserving the causal structure that defines autoregression. To guard against information leakage during training, we introduce a discrete tensor noising mechanism inspired by discrete diffusion theory, which injects categorical noise into input tensors. TensorAR is designed to be plug-and-play: unlike masked AR methods, it requires no architectural modifications, and unlike autoregressive diffusion, it preserves the familiar AR training paradigm. We evaluate TensorAR across both class-to-image and text-to-image tasks, showing consistent gains in generation quality and instruction-following ability, while achieving a superior balance between quality and latency. In doing so, TensorAR offers a new path forward for autoregressive generation---one where predictions are not just produced, but continually refined.

Visual autoregressive (AR) generation offers a promising path toward unifying vision and language models, yet its performance remains suboptimal against diffusion models. Prior work often attributes this gap to tokenizer limitations and rasterization ordering. In this work, we identify a core bottleneck from the perspective of generator-tokenizer inconsistency, i.e., the AR-generated tokens may not be well-decoded by the tokenizer. To address this, we propose reAR, a simple training strategy introducing a token-wise regularization objective: when predicting the next token, the causal transformer is also trained to recover the visual embedding of the current token and predict the embedding of the target token under a noisy context. It requires no changes to the tokenizer, generation order, inference pipeline, or external models. Despite its simplicity, reAR substantially improves performance. On ImageNet, it reduces gFID from 3.02 to 1.86 and improves IS to 316.9 using a standard rasterization-based tokenizer. When applied to advanced tokenizers, it achieves a gFID of 1.42 with only 177M parameters, matching the performance with larger state-of-the-art diffusion models (675M).

Diffusion Models achieve state-of-the-art performance in generating new samples but lack a low-dimensional latent space that encodes the data into editable features. Inversion-based methods address this by reversing the denoising trajectory, transferring images to their approximated starting noise. In this work, we thoroughly analyze this procedure and focus on the relation between the initial noise, the generated samples, and their corresponding latent encodings obtained through the DDIM inversion. First, we show that latents exhibit structural patterns in the form of less diverse noise predicted for smooth image areas (e.g., plain sky). Through a series of analyses, we trace this issue to the first inversion steps, which fail to provide accurate and diverse noise. Consequently, the DDIM inversion space is notably less manipulative than the original noise. We show that prior inversion methods do not fully resolve this issue, but our simple fix, where we replace the first DDIM Inversion steps with a forward diffusion process, successfully decorrelates latent encodings and enables higher quality editions and interpolations.

Diffusion models have transformed image generation, yet controlling their outputs for diverse applications, including content moderation and creative customization, remains challenging. Existing approaches usually require task-specific training and struggle to generalise across both concrete (e.g., objects) and abstract (e.g.,4 styles) concepts. We propose CASteer (Cross-Attention Steering), a training-free framework for controllable image generation using steering vectors to influence a diffusion model’s hidden representations dynamically. CASteer precomputes concept-specific steering vectors by averaging neural activations from images generated for each target concept. During inference, it dynamically applies these vectors to modify outputs only when necessary, either removing undesired concepts from images where they appear or adding desired concepts to images where they are absent. This selective activation ensures precise, context-aware adjustments without altering unaffected regions. This approach enables precise control over a wide range of tasks, including removing harmful content, interpolating between desired attributes, replacing objects, all without model retraining. CASteer outperforms state-of-the-art techniques while preserving unrelated content and minimising unintended effects.

Diffusion transformers typically incorporate textual information via (i) attention layers and (ii) a modulation mechanism using a pooled text embedding. Nevertheless, recent approaches discard modulation-based text conditioning and rely exclusively on attention. In this paper, we address whether modulation-based text conditioning is necessary and whether it can provide any performance advantage. Our analysis shows that, in its conventional usage, the pooled embedding contributes little to overall performance, suggesting that attention alone is generally sufficient for faithfully propagating prompt information. However, we reveal that the pooled embedding can provide significant gains when used from a different perspective—serving as guidance and enabling controllable shifts toward more desirable properties. This approach is training-free, simple to implement, incurs negligible runtime overhead, and can be applied to various diffusion models, bringing improvements across diverse tasks, including text-to-image/video generation and image editing.

Diffusion transformers exhibit remarkable video generation capability, yet their prohibitive computational and memory costs hinder practical deployment. Model quantization and attention sparsification are two promising directions for compression, but each alone suffers severe performance degradation under aggressive compression. Combining them promises compounded efficiency gains, but naive integration is ineffective. The sparsity-induced information loss exacerbates quantization noise, leading to amplified attention shifts. To address this, we propose **QuantSparse**, a unified framework that integrates model quantization with attention sparsification. Specifically, we introduce *Multi-Scale Salient Attention Distillation*, which leverages both global structural guidance and local salient supervision to mitigate quantization-induced bias. In addition, we develop *Second-Order Sparse Attention Reparameterization*, which exploits the temporal stability of second-order residuals to efficiently recover information lost under sparsity. Experiments on HunyuanVideo-13B demonstrate that QuantSparse achieves 20.88 PSNR, substantially outperforming the state-of-the-art quantization baseline Q-VDiT (16.85 PSNR), while simultaneously delivering a **3.68$\times$** reduction in storage and **1.88$\times$** acceleration in end-to-end inference.

Generative models for Image Super-Resolution (SR) are increasingly powerful, yet their reliance on self-attention's quadratic complexity ($O(N^2)$) creates a major computational bottleneck. Linear Attention offers an $O(N)$ solution, but its promise for photorealistic SR has remained largely untapped, historically hindered by a cascade of interrelated and previously unsolved challenges. This paper introduces LinearSR, a holistic framework that, for the first time, systematically overcomes these critical hurdles. Specifically, we resolve a fundamental, training instability that causes catastrophic model divergence using our novel ''knee point''-based Early-Stopping Guided Fine-tuning (ESGF) strategy. Furthermore, we mitigate the classic perception-distortion trade-off with a dedicated SNR-based Mixture of Experts (MoE) architecture. Finally, we establish an effective and lightweight guidance paradigm, TAG, derived from our ''precision-over-volume'' principle. Our resulting LinearSR model simultaneously delivers state-of-the-art perceptual quality with exceptional efficiency. Its core diffusion forward pass (1-NFE) achieves SOTA-level speed, while its overall multi-step inference time remains highly competitive. This work provides the first robust methodology for applying Linear Attention in the photorealistic SR domain, establishing a foundational paradigm for future research in efficient generative super-resolution.

Diffusion Transformers (DiTs) have emerged as the state-of-the-art architecture for video generation, yet their computational and memory demands hinder practical deployment. While post-training quantization (PTQ) presents a promising approach to accelerate Video DiT models, existing methods suffer from two critical limitations: (1) dependence on computation-heavy and inflexible calibration procedures, and (2) considerable performance deterioration after quantization. To address these challenges, we propose DVD-Quant, a novel Data-free quantization framework for Video DiTs. Our approach integrates three key innovations: (1) Bounded-init Grid Refinement (BGR) and (2) Auto-scaling Rotated Quantization (ARQ) for calibration data-free quantization error reduction, as well as (3) $\delta$-Guided Bit Switching ($\delta$-GBS) for adaptive bit-width allocation. Extensive experiments across multiple video generation benchmarks demonstrate that DVD-Quant achieves an approximately 2$\times$ speedup over full-precision baselines on advanced DiT models while maintaining visual fidelity. Notably, DVD-Quant is the first to enable W4A4 PTQ for Video DiTs without compromising video quality. Code and models will be released to facilitate future research.

Personalizing text-to-image diffusion models involves integrating novel visual concepts from a small set of reference images while retaining the model’s original generative capabilities. However, this process often leads to overfitting, where the model ignores the user’s prompt and merely replicates the reference images. We attribute this issue to a fundamental misalignment between the true goals of personalization, which are subject fidelity and text alignment, and the training objectives of existing methods that fail to enforce both objectives simultaneously. Specifically, prior approaches often overlook the need to explicitly preserve the pretrained model’s output distribution, resulting in distributional drift that undermines diversity and coherence. To resolve these challenges, we introduce a Lipschitz-based regularization objective that constrains parameter updates during personalization, ensuring bounded deviation from the original distribution. This promotes consistency with the pretrained model’s behavior while enabling accurate adaptation to new concepts. Furthermore, our method offers a computationally efficient alternative to commonly used, resource-intensive sampling techniques. Through extensive experiments across diverse diffusion model architectures, we demonstrate that our approach achieves superior performance in both quantitative metrics and qualitative evaluations, consistently excelling in visual fidelity and prompt adherence. We further support these findings with comprehensive analyses, including ablation studies and visualizations.

Existing multi-view 3D object reconstruction methods heavily rely on sufficient overlap between input views, where occlusions and sparse coverage in practice frequently yield severe reconstruction incompleteness. Recent advancements in diffusion-based 3D generative techniques offer the potential to address these limitations by leveraging learned generative priors to ‘‘hallucinate" invisible parts of objects, thereby generating plausible 3D structures. However, the stochastic nature of the inference process limits the accuracy and reliability of generation results, preventing existing reconstruction frameworks from integrating such 3D generative priors. In this work, we comprehensively analyze the reasons why diffusion-based 3D generative methods fail to achieve high consistency, including (a) the insufficiency in constructing and leveraging cross-view connections when extracting multi-view image features as conditions, and (b) the poor controllability of iterative denoising during local detail generation, which easily leads to plausible but inconsistent fine geometric and texture details with inputs. Accordingly, we propose ReconViaGen to innovatively integrate reconstruction priors into the generative framework and devise several strategies that effectively address these issues. Extensive experiments demonstrate that our ReconViaGen can reconstruct complete and accurate 3D models consistent with input views in both global structure and local details.

Gaussian Splatting (GS) enables immersive rendering, but realistic 3D object–scene composition remains challenging. Baked appearance and shadow information in GS radiance fields cause inconsistencies when combining objects and scenes. Addressing this requires relightable object reconstruction and scene lighting estimation. For relightable object reconstruction, existing Gaussian-based inverse rendering methods often rely on ray tracing, leading to low efficiency. We introduce Surface Octahedral Probes (SOPs), which store lighting and occlusion information and allow efficient 3D querying via interpolation, avoiding expensive ray tracing. SOPs provide at least a 2x speedup in reconstruction and enable real-time shadow computation in Gaussian scenes. For lighting estimation, existing Gaussian-based inverse rendering methods struggle to model intricate light transport and often fail in complex scenes, while learning-based methods predict lighting from a single image and are viewpoint-sensitive. We observe that 3D object–scene composition primarily concerns the object’s appearance and nearby shadows. Thus, we simplify the challenging task of full scene lighting estimation by focusing on the environment lighting at the object’s placement. Specifically, we capture a 360° reconstructed radiance field of the scene at the location and fine-tune a diffusion model to complete the lighting. Building on these advances, we propose ComGS, a novel 3D object–scene composition framework. Our method achieves high-quality, real-time rendering at around 26 FPS, produces visually harmonious results with vivid shadows, and requires only 36 seconds for editing. The code and dataset are available at https://nju-3dv.github.io/projects/ComGS/.

Robust 3D hand reconstruction is challenging in egocentric vision due to depth ambiguity, self-occlusion, and complex hand-object interactions. Prior works attempt to mitigate the challenges by scaling up training data or incorporating auxiliary cues, often falling short of effectively handling unseen contexts. In this paper, we introduce EgoHandICL, the first in-context learning (ICL) framework for 3D hand reconstruction that achieves strong semantic alignment, visual consistency, and robustness under challenging egocentric conditions. Specifically, we develop (i) complementary exemplar retrieval strategies guided by vision–language models (VLMs), (ii) an ICL-tailored tokenizer that integrates multimodal context, and (iii) a Masked Autoencoders (MAE)-based architecture trained with 3D hand–guided geometric and perceptual objectives. By conducting comprehensive experiments on the ARCTIC and EgoExo4D benchmarks, our EgoHandICL consistently demonstrates significant improvements over state-of-the-art 3D hand reconstruction methods. We further show EgoHandICL’s applicability by testing it on real-world egocentric cases and integrating it with EgoVLMs to enhance their hand–object interaction reasoning. Our code and data will be publicly available.

The recognition of dynamic and social behavior in animals is fundamental for advancing several areas of the life sciences, including ethology, ecology, medicine and neuroscience. Recent progress in deep learning has enabled an automated recognition of such behavior from video data. However, an accurate reconstruction of the three-dimensional (3D) pose and shape has not been integrated into this process. Especially for non-human primates, the animals phylogenetically closest to humans, mesh-based tracking efforts lag behind those for other species, leaving pose descriptions restricted to sparse keypoints that are unable to fully capture the richness of action dynamics. To address this gap, we introduce the $\textbf{Big Ma}$ca$\textbf{Q}$ue 3D Motion and Animation Dataset ($\texttt{BigMaQ}$), a large-scale dataset comprising more than 750 scenes of interacting rhesus macaques with detailed 3D pose descriptions of skeletal joint rotations. Recordings were obtained from 16 calibrated cameras and paired with action labels derived from a curated ethogram. Extending previous surface-based animal tracking methods, we construct subject-specific textured avatars by adapting a high-quality macaque template mesh to individual monkeys. This allows us to provide pose descriptions that are more accurate than previous state-of-the-art surface-based animal tracking methods. From the original dataset, we derive BigMaQ500, an action recognition benchmark that links surface-based pose vectors to single frames across multiple individual monkeys. By pairing features extracted from established image and video encoders with and without our pose descriptors, we demonstrate substantial improvements in mean average precision (mAP) when pose information is included. With these contributions, $\texttt{BigMaQ}$ establishes the first dataset that both integrates dynamic 3D pose-shape representations into the learning task of animal action recognition and provides a rich resource to advance the study of visual appearance, posture, and social interaction in non-human primates. The code and data are publicly available at [https://martinivis.github.io/BigMaQ/](https://martinivis.github.io/BigMaQ/).

Single-view 3D human reconstruction has achieved remarkable progress through the adoption of multi-view diffusion models, yet the recovered 3D humans often exhibit unnatural poses. This phenomenon becomes pronounced when reconstructing 3D humans with dynamic or challenging poses, which we attribute to the limited scale of available 3D human datasets with diverse poses. To address this limitation, we introduce DrPose, Direct Reward fine-tuning algorithm on Poses, which enables post-training of a multi-view diffusion model on diverse poses without requiring expensive 3D human assets. DrPose trains a model using only human poses paired with single-view images, employing a direct reward fine-tuning to maximize PoseScore, which is our proposed differentiable reward that quantifies consistency between a generated multi-view latent image and a ground-truth human pose. This optimization is conducted on DrPose15K, a novel dataset that was constructed from an existing human motion dataset and a pose-conditioned video generative model. Constructed from abundant human pose sequence data, DrPose15K exhibits a broader pose distribution compared to existing 3D human datasets. We validate our approach through evaluation on conventional benchmark datasets, in-the-wild images, and a newly constructed benchmark, with a particular focus on assessing performance on challenging human poses. Our results demonstrate consistent qualitative and quantitative improvements across all benchmarks. Project page: https://seunguk-do.github.io/drpose.

Textured 3D morphing seeks to generate smooth and plausible transitions between two 3D assets, preserving both structural coherence and fine-grained appearance. This ability is crucial not only for advancing 3D generation research but also for practical applications in animation, editing, and digital content creation. Existing approaches either operate directly on geometry, limiting them to shape-only morphing while neglecting textures, or extend 2D interpolation strategies into 3D, which often causes semantic ambiguity, structural misalignment, and texture blurring. These challenges underscore the necessity to jointly preserve geometric consistency, texture alignment, and robustness throughout the transition process. To address this, we propose Interp3D, a novel training-free framework for textured 3D morphing. It harnesses generative priors and adopts a progressive alignment principle to ensure both geometric fidelity and texture coherence. Starting from semantically aligned interpolation in condition space, Interp3D enforces structural consistency via SLAT (Structured Latent)-guided structure interpolation, and finally transfers appearance details through fine-grained texture fusion. For comprehensive evaluations, we construct a dedicated dataset, Interp3DData, with graded difficulty levels and assess generation results from fidelity, transition smoothness, and plausibility. Both quantitative metrics and human studies demonstrate the significant advantages of our proposed approach over previous methods. Source code is available at https://github.com/xiaolul2/Interp3D.

Recent advances in dynamic scene reconstruction have significantly benefited from 3D Gaussian Splatting, yet existing methods show inconsistent performance across diverse scenes, indicating no single approach effectively handles all dynamic challenges. To overcome these limitations, we propose Mixture of Experts for Dynamic Gaussian Splatting (MoE-GS), a unified framework integrating multiple specialized experts via a novel Volume-aware Pixel Router. Unlike sparsity-oriented MoE architectures in large language models, MoE-GS is designed to improve dynamic novel view synthesis quality by combining heterogeneous deformation priors, rather than to reduce training or inference-time FLOPs. Our router adaptively blends expert outputs by projecting volumetric Gaussian-level weights into pixel space through differentiable weight splatting, ensuring spatially and temporally coherent results. Although MoE-GS improves rendering quality, the increased model capacity and reduced FPS are inherent to the MoE architecture. To mitigate this, we explore two complementary directions: (1) single-pass multi-expert rendering and gate-aware Gaussian pruning, which improve efficiency within the MoE framework, and (2) a distillation strategy that transfers MoE performance to individual experts, enabling lightweight deployment without architectural changes. To the best of our knowledge, MoE-GS is the first approach incorporating Mixture-of-Experts techniques into dynamic Gaussian splatting. Extensive experiments on the N3V and Technicolor datasets demonstrate that MoE-GS consistently outperforms state-of-the-art methods with improved efficiency. Video demonstrations are available at cvsp-lab.github.io/MoE-GS.

Reconstruction techniques, such as 3D Gaussian Splatting (3DGS), are increasingly used to generate scenarios for autonomous driving system (ADS) research. Existing 3DGS-based approaches for autonomous-driving scenario generation have, through various optimizations, achieved high visual similarity in reconstructed scenes. However, this route is built on a strong assumption: that higher scene similarity directly translates into better preservation of ADS behaviour. Unfortunately, this assumption has not been effectively validated, and ADS behaviour is more closely related to objects within the field of view rather than the global image. Thus, we focus on the perception module—the entry point of ADS. Preliminary experiments reveal that although current methods can produce reconstructions with high overall similarity, they often fail to ensure that the perception module outputs remain consistent with those obtained from the original images. Such a limitation can significantly harm the applicability of reconstruction in the ADS domain. To address this gap, we propose two complementary solutions: a perception-aligned loss, which directly leverages output differences between reconstructed and ground-truth images during training; and an object zone quality loss, which specifically reinforces training on object locations identified by the perception model on ground-truth images. Experiments demonstrate that both of our methods improve the ability of reconstructed scenes to maintain consistency between the perception module outputs and the ground-truth inputs. We release code at: https://github.com/Shanicky-RenzhiWang/Perception-aware-3DGS

Surfaces are typically represented as meshes, which can be extracted from volumetric fields via meshing or optimized directly as surface parameterizations. Volumetric representations occupy 3D space and have a large effective receptive field along rays, enabling stable and efficient optimization via volumetric rendering; however, subsequent meshing often produces overly dense meshes and introduces accumulated errors. In contrast, pure surface methods avoid meshing but capture only boundary geometry with a single-layer receptive field, making it difficult to learn intricate geometric details and increasing reliance on priors (e.g., shading or normals). We bridge this gap by differentiably turning a surface representation into a volumetric one, enabling end-to-end surface reconstruction via volumetric rendering to model complex geometries. Specifically, we soften a mesh into multiple semi-transparent layers that remain differentiable with respect to the base mesh, endowing it with a controllable 3D receptive field. Combined with a splatting-based renderer and a topology-control strategy, our method can be optimized in about 20 minutes to achieve accurate surface reconstruction while substantially improving mesh quality.

Implicit Neural Representations (INRs) have emerged as promising surrogates for large 3D scientific simulations due to their ability to continuously model spatial and conditional fields, yet they face a critical fidelity-speed dilemma: deep MLPs suffer from high inference cost, while efficient embedding-based models lack sufficient expressiveness. To resolve this, we propose the Decoupled Representation Refinement (DRR) architectural paradigm. DRR leverages a deep refiner network, alongside non-parametric transformations, in a one-time offline process to encode rich representations into a compact and efficient embedding structure. This approach decouples slow neural networks with high representational capacity from the fast inference path. We introduce DRR-Net, a simple network that validates this paradigm, and a novel data augmentation strategy, Variational Pairs (VP) for improving INRs under complex tasks like high-dimensional surrogate modeling. Experiments on several ensemble simulation datasets demonstrate that our approach achieves state-of-the-art fidelity, while being up to 27$\times$ faster at inference than high-fidelity baselines and remaining competitive with the fastest models. The DRR paradigm offers an effective strategy for building powerful and practical neural field surrogates and INRs in broader applications, with a minimal compromise between speed and quality.

While current high-resolution depth estimation methods achieve strong results, they often suffer from computational inefficiencies due to reliance on heavyweight models and multiple inference steps, increasing inference time. To address this, we introduce PatchRefiner V2 (PRV2), which replaces heavy refiner models with lightweight encoders. This reduces model size and inference time but introduces noisy features. To overcome this, we propose a Coarse-to-Fine (C2F) module with a Guided Denoising Unit for refining and denoising the refiner features and a Noisy Pretraining strategy to pretrain the refiner branch to fully exploit the potential of the lightweight refiner branch. Additionally, we propose to adopt the Scale-and-Shift Invariant Gradient Matching (SSIGM) loss within local windows to enhance synthetic-to-real domain transfer. PRV2 outperforms state-of-the-art depth estimation methods on UnrealStereo4K in both accuracy and speed, using fewer parameters and faster inference. It also shows improved depth boundary delineation on real-world datasets like CityScapes, demonstrating its effectiveness.

We present CogniMap3D, a bioinspired framework for dynamic 3D scene understanding and reconstruction that emulates human cognitive processes. Our approach maintains a persistent memory bank of static scenes, enabling efficient spatial knowledge storage and rapid retrieval. CogniMap3D integrates three core capabilities: a multi-stage motion cue framework for identifying dynamic objects, a cognitive mapping system for storing, recalling, and updating static scenes across multiple visits, and a factor graph optimization strategy for refining camera poses. Given an image stream, our model identifies dynamic regions through motion cues with depth and camera pose priors, then matches static elements against its memory bank. When revisiting familiar locations, CogniMap3D retrieves stored scenes, relocates cameras, and updates memory with new observations. Evaluations on video depth estimation, camera pose reconstruction, and 3D mapping tasks demonstrate its state-of-the-art performance, while effectively supporting continuous scene understanding across extended sequences and multiple visits.

This work enables everyday devices, e.g., smartphones, to dynamically capture open-ended 3D scenes with rich, expandable semantics for immersive virtual worlds. While 3DGS and foundation models hold promise for semantic scene understanding, existing solutions suffer from unscalable semantic integration, prohibitive memory costs, and cross-view inconsistency. To respond, we propose Open-Set Semantic Gaussian Splatting SLAM, a GS-SLAM system augmented by an expandable semantic feature pool that decouples condensed scene-level semantics from individual 3D Gaussians. Each Gaussian references semantics via a lightweight indexing vector, reducing memory overhead by orders of magnitude while supporting dynamic updates. Besides, we introduce a consistency-aware optimization strategy alongside a Semantic Stability Guidance mechanism to enhance long-term, cross-view semantic consistency and resolve inconsistencies. Experiments demonstrate that our system achieves high-fidelity rendering with scalable, open-set semantics across both controlled and in-the-wild environments, supporting applications like 3D localization and scene editing. These results mark an initial yet solid step towards high-quality, expressive, and accessible 3D virtual world modeling. Our code will be publicly released.

The field of 4D world modeling—aiming to jointly capture spatial geometry and temporal dynamics—has witnessed remarkable progress in recent years, driven by advances in large-scale generative models and multimodal learning. However, the development of truly general 4D world models remains fundamentally constrained by the availability of high-quality data. Existing datasets and benchmarks often lack the dynamic complexity, multi-domain diversity, and spatial-temporal annotations required to support key tasks such as 4D geometric reconstruction, future prediction, and camera-controlled video generation. To address this gap, we introduce OmniWorld, a large-scale, multi-domain, multi-modal dataset specifically designed for 4D world modeling. OmniWorld consists of a newly collected OmniWorld-Game dataset and several curated public datasets spanning diverse domains. Compared with existing synthetic datasets, OmniWorld-Game provides richer modality coverage, larger scale, and more realistic dynamic interactions. Based on this dataset, we establish a challenging benchmark that exposes the limitations of current state-of-the-art (SOTA) approaches in modeling complex 4D environments. Moreover, fine-tuning existing SOTA methods on OmniWorld leads to significant performance gains across 4D reconstruction and video generation tasks, strongly validating OmniWorld as a powerful resource for training and evaluation. We envision OmniWorld as a catalyst for accelerating the development of general-purpose 4D world models, ultimately advancing machines’ holistic understanding of the physical world.

Scene understanding with free-form language has been widely explored within diverse modalities such as images, point clouds, and LiDAR. However, related studies on event sensors are scarce or narrowly centered on semantic-level understanding. We introduce SEAL, the first Semantic-aware Segment Any Events framework that addresses Open-Vocabulary Event Instance Segmentation (OV-EIS). Given the visual prompt, our model presents a unified framework to support both event segmentation and open-vocabulary mask classification at multiple levels of granularity, including instance-level and part-level. To enable thorough evaluation on OV-EIS, we curate four benchmarks that cover label granularity from coarse to fine class configurations and semantic granularity from instance-level to part-level understanding. Extensive experiments show that our SEAL largely outperforms proposed baselines in terms of performance and inference speed with a parameter-efficient architecture. In the Appendix, we further present a simple variant of our SEAL achieving generic spatiotemporal OV-EIS that does not require any visual prompts from users in the inference. The code will be publicly available.

The emergence of 3D Gaussian Splatting (3DGS) has significantly advanced Novel View Synthesis (NVS) through explicit scene representation, enabling real-time photorealistic rendering. However, existing approaches manifest two critical limitations in complex scenarios: (1) Over-reconstruction occurs when persistent large Gaussians cannot meet adaptive splitting thresholds during density control. This is exacerbated by conflicting gradient directions that prevent effective splitting of these Gaussians; (2) Over-densification of Gaussians occurs in regions with aligned gradient aggregation, leading to redundant component proliferation. This redundancy significantly increases memory overhead due to unnecessary data retention. We present Gradient-Direction-Aware Gaussian Splatting (GDAGS) to address these challenges. Our key innovations: the Gradient Coherence Ratio (GCR), computed through normalized gradient vector norms, which explicitly discriminates Gaussians with concordant versus conflicting gradient directions; and a nonlinear dynamic weighting mechanism leverages the GCR to enable gradient-direction-aware density control. Specifically, GDAGS prioritizes conflicting-gradient Gaussians during splitting operations to enhance geometric details while suppressing redundant concordant-direction Gaussians. Conversely, in cloning processes, GDAGS promotes concordant-direction Gaussian densification for structural completion while preventing conflicting-direction Gaussian overpopulation. Comprehensive evaluations across diverse real-world benchmarks demonstrate that GDAGS achieves superior rendering quality while effectively mitigating over-reconstruction, suppressing over-densification, and constructing compact scene representations.

Video understanding in multimodal language models remains limited by context length: models often miss key transition frames and struggle to maintain coherence across long time scales. To address this, we adapt Native Sparse Attention (NSA) to video-language models. Our method, VideoNSA, adapts Qwen2.5-VL through end-to-end training on a 216K video instruction dataset. We employ a hardware-aware hybrid approach to attention, preserving dense attention for text, while employing NSA for video. Compared to token-compression and training-free sparse baselines, VideoNSA achieves improved performance on long-video understanding, temporal reasoning, and spatial benchmarks. Further ablation analysis reveals four key findings: (1) reliable scaling to 128K tokens; (2) an optimal global–local attention allocation at a fixed budget; (3) task-dependent branch usage patterns; and (4) the learnable combined sparse attention help induce dynamic attention sinks.

Large Vision-Language Models excel at multimodal understanding but struggle to deeply integrate visual information into their predominantly text-based reasoning processes, a key challenge in mirroring human cognition. To address this, we introduce \nameshort{}, a model that learns to ``think with images'', trained end-to-end with reinforcement learning without requiring pre-collected reasoning data for cold-start supervised fine-tuning (SFT). Notably, this ability emerges natively, leveraging the model's own grounding capability as an intrinsic function rather than relying on external specialized models or APIs. We enable this capability through active perception, where the model learns to strategically ground its reasoning in visual information, guided by a tailored data selection and reward strategy. \nameshort{} achieves significant performance gains on general perception and reasoning benchmarks and also demonstrates improvement in grounding, hallucination, and mathematical reasoning tasks. Interestingly, we observe the distinct evolution of active perception from initial exploration to efficient and accurate exploitation, and diverse thinking patterns that closely mirror human visual reasoning processes. Code is available at \url{https://github.com/Visual-Agent/DeepEyes}.

Multimodal large language models (MLLMs) have advanced rapidly in recent years. However, existing approaches for vision tasks often rely on indirect representations, such as generating coordinates as text for detection, which limits performance and prevents dense prediction tasks like segmentation. To overcome these challenges, we introduce Patch-as-Decodable Token (PaDT), a unified paradigm that enables MLLMs to directly generate both textual and diverse visual outputs. Central to PaDT are Visual Reference Tokens (VRTs), derived from visual patch embeddings of query images and interleaved seamlessly with LLM's output textual tokens. A lightweight decoder then transforms LLM's outputs into detection, segmentation, and grounding predictions. Unlike prior methods, PaDT processes VRTs independently at each forward pass and dynamically expands the embedding table, thus improving localization and differentiation among similar objects. We further tailor a training strategy for PaDT by randomly selecting VRTs for supervised fine-tuning and introducing a robust per-token cross-entropy loss. Our empirical studies across four visual perception and understanding tasks suggest PaDT consistently achieving state-of-the-art performance, even compared with significantly larger MLLM models. The code is available at https://github.com/Gorilla-Lab-SCUT/PaDT.

Recent advancements in Large Language Models (LLMs) and their multimodal extensions (MLLMs) have substantially enhanced machine reasoning across diverse tasks. However, these models predominantly rely on pure text as the medium for both expressing and structuring reasoning, even when visual information is present. In this work, we argue that language may not always be the most natural or effective modality for reasoning, particularly in tasks involving spatial and geometrical information. Motivated by this, we propose a new paradigm, Visual Planning, which enables planning through purely visual representations for these "vision-first'' tasks, as a supplementary channel to language-based reasoning. In this paradigm, planning is executed via sequences of images that encode step-by-step inference in the visual domain, akin to how humans sketch or visualize future actions. We introduce a novel reinforcement learning framework, Visual Planning via Reinforcement Learning (VPRL), empowered by GRPO for post-training large vision models, leading to substantial improvements in planning in a selection of representative visual navigation tasks, FrozenLake, Maze, and MiniBehavior. Our visual planning paradigm outperforms all other planning variants that conduct reasoning in the text-only space. Our results establish Visual Planning as a viable and promising supplement to language-based reasoning, opening new avenues for tasks that benefit from intuitive, image-based inference.

Diffusion language models enable any-order generation and bidirectional conditioning, offering appealing flexibility for tasks such as infilling, rewriting, and self-correction. However, their formulation—predicting one part of a sequence from another within a single-step dependency—limits modeling depth and often yields lower sample quality and stability than autoregressive (AR) models. To address this, we revisit autoregressive modeling as a foundation and reformulate diffusion-style training into a structured multi-group prediction process. We propose Any-order Any-subset Autoregressive modeling (A3), a generalized framework that extends the standard AR factorization to arbitrary token groups and generation orders. A3 preserves the probabilistic rigor and multi-layer dependency modeling of AR while inheriting diffusion models' flexibility for parallel and bidirectional generation. We implement A3 through a two-stream attention architecture and a progressive adaptation strategy that transitions pretrained AR models toward any-order prediction. Experiments on question answering, commonsense reasoning, and story infilling demonstrate that A3 outperforms diffusion-based models while maintaining flexible decoding. This work offers a unified approach for a flexible, efficient, and novel language modeling paradigm. Code is at https://github.com/PKU-ML/Any-order-Any-subset-AR.

Audiovisual video captioning aims to generate semantically rich descriptions with temporal alignment between visual and auditory events, thereby benefiting both video understanding and generation. In this paper, we present AVoCaDO, a powerful audiovisual video captioner driven by the temporal orchestration between audio and visual modalities. We propose a two-stage post-training pipeline: (1) AVoCaDO SFT, which fine-tunes the model on a newly curated dataset of 107K high-quality, temporally-aligned audiovisual captions; and (2) AVoCaDO GRPO, which leverages tailored reward functions to further enhance temporal coherence and dialogue accuracy while regularizing caption length and reducing collapse. Experimental results demonstrate that AVoCaDO significantly outperforms existing open-source models across four audiovisual video captioning benchmarks, and also achieves competitive performance on the VDC benchmark under visual-only settings. The model will be made publicly available to facilitate future research in audiovisual video understanding and generation.

Temporal grounding in long, untrimmed videos is critical for real-world video understanding, yet it remains a challenging task owing to complex temporal structures and pervasive visual redundancy. Existing methods rely heavily on supervised training with task-specific annotations, which inherently limits their scalability and adaptability due to the substantial cost of data collection and model retraining. Although a few recent works have explored training-free or zero-shot grounding, they seldom address the unique challenges posed by long videos. In this paper, we propose HiTeA (Hierarchical Temporal Alignment), a novel, training-free framework explicitly designed for long-video temporal grounding. HiTeA introduces a hierarchical temporal decomposition mechanism that structures videos into events, scenes, and actions, thereby aligning natural language queries with the most appropriate temporal granularity. Candidate segments are then matched with queries by leveraging pre-trained vision–language models (VLMs) to directly compute segment–text similarity, thereby obviating the need for any task-specific training or fine-tuning. Extensive experiments on both short- and long-video benchmarks show that HiTeA not only substantially outperforms all existing training-free methods (e.g., achieving 44.94% R\@0.1 on TACoS, representing an absolute gain of 12.4%) but also achieves competitive performance against state-of-the-art supervised baselines under stricter metrics. The code is available at https://anonymous.4open.science/r/HiTeA_code.

Understanding the interplay between intra-modality dependencies (the contribution of an individual modality to a target task) and inter-modality dependencies (the relationships between modalities and the target task) is fundamental to advancing multi-modal learning. However, the nature of and interaction between these dependencies within current benchmark evaluations remains poorly characterized. In this work, we present a large-scale empirical study to quantify these dependencies across 23 visual question-answering benchmarks using multi-modal large language models (MLLMs) covering domains such as general and expert knowledge reasoning, optical character recognition, and document understanding. Our findings show that the reliance on vision, question (text), and their interaction varies significantly, both across and within benchmarks. We discover that numerous benchmarks intended to mitigate text-only biases have inadvertently amplified image-only dependencies. This characterization persists across model sizes and types, with models often obtaining high performance by using each modality independently and showing limited dependence on their interaction. We provide a quantitative characterization of multi-modal datasets, enabling a principled approach to multi-modal benchmark design and evaluation.

Audio and music generation based on flexible multimodal control signals is a widely applicable topic, with the following key challenges: 1) a unified multimodal modeling framework, and 2) large-scale, high-quality training data. As such, we propose AudioX, a unified framework for anything-to-audio generation that integrates varied multimodal conditions (i.e., text, video, image, and audio signals) in this work. The core design in this framework is a Multimodal Adaptive Fusion module, which enables the effective fusion of diverse multimodal inputs, enhancing cross-modal alignment and improving overall generation quality. To train this unified model, we construct a large-scale, high-quality dataset, IF-caps, comprising over 7 million samples curated through a structured data annotation pipeline. This dataset provides comprehensive supervision for multimodal-conditioned audio generation. We benchmark AudioX against state-of-the-art methods across a wide range of tasks, finding that our model achieves superior performance, especially in text-to-audio and text-to-music generation. These results demonstrate our method is capable of audio generation under multimodal control signals, showing powerful instruction-following potential. We will release the code, model, and dataset.

Reinforcement learning (RL) has shown strong potential for enhancing reasoning in multimodal large language models (MLLMs), yet existing video reasoning methods often rely on coarse sequence-level rewards or single-factor token selection. Such approaches neglect fine-grained links among visual inputs, temporal dynamics, and linguistic outputs, limiting both accuracy and interpretability. We propose Video-KTR, a modality-aware policy shaping framework that performs selective, token-level RL by combining three attribution signals: (1) visual-aware tokens identified via counterfactual masking to reveal perceptual dependence; (2) temporal-aware tokens detected through frame shuffling to expose causal and temporal sensitivity; and (3) high-entropy tokens signaling predictive uncertainty. By reinforcing only the union of key tokens, Video-KTR focuses learning on semantically informative, modality-sensitive content while filtering out low-value tokens. Across five challenging benchmarks, Video-KTR achieves state-of-the-art or highly competitive results—42.7% on Video-Holmes, surpassing GPT-4o—with consistent gains on both reasoning-centric and general video understanding tasks. Ablation studies verify the complementary roles of the attribution signals and the robustness of targeted token-level updates. Overall, Video-KTR improves accuracy and interpretability, offering a simple, drop-in extension to RL for complex video reasoning.

Significant progress has been made in reasoning segmentation by combining multi-modal large language models (MLLMs) with the Segment Anything Model (SAM): the former excel in reasoning and vision–language alignment, while the latter offers powerful pixel-level understanding. However, current paradigms fall short in exploiting SAM’s strengths, especially the ability to support iterative mask refinement by interactive segmentation, a process that human users can naturally perform. To bridge this gap, we introduce SAM-Veteran, an experienced mask-aware SAM agent capable of emulating human interaction with SAM via a reasoning-driven segmentation workflow that integrates (i) generating bounding boxes given image–query pairs for SAM input, (ii) proposing refinement points based on SAM-generated masks, and (iii) adaptively terminating the process. Aiming for this goal, we propose a multi-task reinforcement learning framework based on Group Relative Policy Optimization (GRPO), which enhances the MLLM’s abilities in textual grounding and mask comprehension. Furthermore, we introduce a dynamic sampling strategy tailored for generating both boxes and points to stabilize training. Extensive experiments across diverse datasets show that SAM-Veteran achieves human-like interaction with SAM and establishes new state-of-the-art performance on both in-domain and out-of-domain benchmarks.

Spine disorders affect 619 million people globally and are a leading cause of disability, yet AI-assisted diagnosis remains limited by the lack of level-aware, multimodal datasets. Clinical decision-making for spine disorders requires sophisticated reasoning across X-ray, CT, and MRI at specific vertebral levels. However, progress has been constrained by the absence of traceable, clinically-grounded instruction data and standardized, spine-specific benchmarks. To address this, we introduce SpineMed, an ecosystem co-designed with practicing spine surgeons. It features SpineMed-450k, the first large-scale dataset explicitly designed for vertebral-level reasoning across imaging modalities with over 450,000 instruction instances, and SpineBench, a clinically-grounded evaluation framework. SpineMed-450k is curated from diverse sources, including textbooks, guidelines, open datasets, and $\sim$1,000 de-identified hospital cases, using a clinician-in-the-loop pipeline with a two-stage LLM generation method (draft and revision) to ensure high-quality, traceable data for question-answering, multi-turn consultations, and report generation. SpineBench evaluates models on clinically salient axes, including level identification, pathology assessment, and surgical planning. Our comprehensive evaluation of several recently advanced large vision-language models (LVLMs) on SpineBench reveals systematic weaknesses in fine-grained, level-specific reasoning. In contrast, our model fine-tuned on SpineMed-450k demonstrates consistent and significant improvements across all tasks. Clinician assessments confirm the diagnostic clarity and practical utility of our model's outputs.

ViTs are general and accurate, and address many tasks, but ViTs are slow, and are not always practical when efficiency is key. Existing methods for faster ViTs design hybrid non-ViT architectures, losing generality, or shrink their tokens, sacrificing accuracy. Many non-ViT architectures are both fast and accurate. Yet, without significant modifications, they cannot do what ViTs can: process other input shapes, pre-train by SOTA self-supervised learning, reduce computation by dropping tokens, and more. We make ViTs faster by reducing patch token width while increasing global token width by adding a new Jumbo token. Our wider Jumbo token is processed by its own wider FFN to increase model capacity. Yet our Jumbo FFN is efficient: it processes a single token, for speed, and its parameters are shared across all layers, for memory. Crucially, our Jumbo is attention-only and non-hierarchical, like a plain ViT, so it is simple, scalable, flexible, and compatible with ViT methods new and old. Jumbo improves over ViT baselines with Registers from Nano to Large scales while maintaining speed/throughput on ImageNet-1K (0.1-13%). Jumbo also improves segmentation (1.9-3.1% on ADE20K), MAE pre-training (4.9% linear probing on ImageNet-1K), test-time adaptation (5.2% on ImageNet-C), and time series modeling. Our Jumbo models even achieve better speed-accuracy trade-offs than specialized non-ViT compute-efficient models, while maintaining plain-ViT compatibility for practicality. Code and weights are available: https://github.com/antofuller/jumbo

Various vision intrusion detection models have achieved great success in many scenarios, e.g., autonomous driving, intelligent monitoring, and security, etc. However, their reliance on pre-defined classes limits their applicability in open-world intrusion detection scenarios. To remedy these, we introduce the Open-Vocabulary Intrusion Detection (OVID) project for the first time. Specifically, we first develop a novel dataset, Cityintrusion-OpenV for OVID, with more diverse intrusion categories and corresponding text prompts. Then, we design a multi-modal, multi-task, and end-to-end open-vocabulary intrusion detection framework named OVIDNet. It achieves open-world intrusion detection via aligning visual features with language embeddings. Further, two simple yet effective strategies are proposed to improve the generalization and performance of this specific task: (1) A Multi-Distributed Noise Mixing strategy is introduced to enhance the location information of unknown and unseen categories. (2) A Dynamic Memory-Gated module is designed to capture the contextual information under complex scenarios. Finally, comprehensive experiments and comparisons are conducted on multiple dominant datasets, e.g., COCO, Cityscape, Foggy-Cityscape, and Cityintrusion-OpenV. Besides, we also evaluate the universal applicability of our model in real scenarios. The results show that our method can outperform other classic and promising methods, and reach strong performance even under task-specific transfer and zero-shot settings, demonstrating its high practicality.

App agents, which autonomously operate mobile Apps through GUIs, have gained significant interest in real-world applications. Yet, they often struggle with long-horizon planning, failing to find the optimal actions for complex tasks with longer steps. To address this, world models are used to predict the next GUI observation based on user actions, enabling more effective agent planning. However, existing world models primarily focus on generating only textual descriptions, lacking essential visual details. To fill this gap, we propose ViMo, the first Visual world Model designed to generate future App observations as images. For the challenge of generating text in image patches, where even minor pixel errors can distort readability, we decompose GUI generation into graphic and text content generation. We propose a novel data representation, the Symbolic Text Representation (STR), to overlay text content with symbolic placeholders while preserving graphics. With this design, ViMo employs a STR Predictor to predict future GUIs’ graphics and a GUI-text Predictor for generating the corresponding text. Moreover, we deploy ViMo to enhance agent-focused tasks by predicting the outcome of actions. Experiments show that ViMo establishes visual world models as a compelling alternative to language-based approaches, producing visually plausible and functionally effective GUIs that empower App agents with more informed decisions.

Multimodal Large Language Models (MLLMs) have recently emerged as general architectures capable of reasoning over diverse modalities. Benchmarks for MLLMs should measure their ability for cross‑modal integration. However, current benchmarks are filled with shortcut questions, which can be solved using only single modality, and thereby yielding unreliable rankings. For example, in vision-language cases, we can find the correct answer without either the image or the text. These low-quality questions unnecessarily increase the size and computational requirements of benchmarks. We introduce a multi-modal and multidimensional item response theory framework (M3IRT) that extends classical IRT by decomposing both model ability and item difficulty into image‑only, text‑only, and cross‑modal components. M3IRT estimates cross‑modal ability of MLLMs and each question’s cross‑modal difficulty, enabling compact, high‑quality subsets that better reflect multimodal reasoning. Across 24 VLMs on three benchmarks, M3IRT prioritizes genuinely cross‑modal questions over shortcuts and preserves ranking fidelity even when 50\% of items are artificially generated low‑quality questions, thereby reducing evaluation cost while improving reliability. M3IRT thus offers a practical tool for assessing cross‑modal reasoning and refining multimodal benchmarks.

Information Bottleneck (IB) Theory is renowned for its ability to learn simple, compact, and effective data representations. In multi-modal clustering, IB theory effectively eliminates interfering redundancy and noise from multi-modal data, while maximally preserving the discriminative information. Existing IB-based multi-modal clustering methods suffer from low-quality pseudo-labels and over-reliance on accurate Mutual Information (MI) estimation, which is known to be challenging. Moreover, unreliable or noisy pseudo-labels may lead to an overconfident clustering outcome. To address these challenges, this paper proposes a novel CaLibrated Information Bottleneck (CLIB) framework designed to learn a clustering that is both accurate and trustworthy. We build a parallel multi-head network architecture—incorporating one primary cluster head and several modality-specific calibration heads—which achieves three key goals: namely, calibrating for the distortions introduced by biased MI estimation thus improving the stability of IB, constructing reliable target variables for IB from multiple modalities and producing a trustworthy clustering result. Notably, we design a dynamic pseudo-label selection strategy based on information redundancy theory to extract high-quality pseudo-labels, thereby enhancing training stability. Experimental results demonstrate that our model not only achieves competitive clustering accuracy on multiple benchmark datasets but also exhibits excellent performance on the expected calibration error metric. Code is available at \textcolor{red}{https://shizhehu.github.io/}.

Conventional wisdom in selecting supervision data for multimodal large language models (MLLMs) is to prioritize datasets that are intuitively similar to the target task (e.g. text-rich v.s. vision-centric). However, it remains unclear how reliably such similarity translates into improved performance on the test benchmarks. In this paper, we take the first step to study the problem in MLLMs: can we predict a training data's influence on a target benchmark even before any training takes place? To answer this question, we first conduct an in-depth analysis using 14 vision-language datasets covering 7 diverse tasks. Our analysis shows that intuitive task similarity is unreliable in predicting task generalizability, and that transfer depends on the specific dataset rather than the broader task category. We propose DATAPROPHET, a training-free, simple yet effective metric based on multimodal perplexity, similarity, and data diversity. Our experiments demonstrate that the influence rankings for different supervision datasets derived from DATAPROPHET is strongly-correlated with rankings based on the actual performance increase after training, with a Kendall’s $\tau$ correlation coefficient of 86.0\%. Moreover, we show that DATAPROPHET can help select better supervision data, achieving up to 6.9\% improvement in average over uniform selection, 1.4\% over SoTA training-based baseline, and 0.2\% higher than oracle experiment performance-based selection. Our code and data will be released.

3D few-shot class-incremental learning (3D FSCIL) requires effectively integrating novel classes from limited samples while preserving base-class knowledge, without succumbing to catastrophic forgetting the learned knowledge or overfitting the novel ones. Current 3D FSCIL approaches predominantly focus on fine-tuning feature representations yet retain static decision boundaries. This leads to a critical trade-off: excessive adaptation to new samples tends to erase previously learned knowledge, while insufficient adaptation hinders novel-class recognition. We argue that the key to effective incremental learning lies not only in feature enhancement but also in adaptive decision-making. To this end, we introduce Point-UQ, an incremental training-free paradigm for 3D point clouds based on uncertainty quantification, which shifts the focus from feature tuning to dynamic decision optimization. Point-UQ comprises two co-designed modules: Attention-driven Adaptive Enhancement (AAE) and Uncertainty-quantification Decision Decoupling (UDD). The former module fuses multi-scale features into calibrated representations, where prediction entropy serves as a reliable measure of per-sample epistemic uncertainty while preserving original feature semantics. Building on AAE-derived calibrated entropy, the UDD module dynamically arbitrates between semantic classifiers and geometric prototypes—enabling robust base-class knowledge retention and accurate novel-class recognition in 3D FSCIL without retraining. Extensive experiments on ModelNet, ShapeNet, ScanObjectNN, and CO3D demonstrate that our approach outperforms state-of-the-art methods by 4% in average accuracy, setting a new standard for robust 3D incremental learning.

Spatial cognition is essential for human intelligence, enabling problem-solving through visual simulations rather than relying solely on verbal reasoning. However, existing AI benchmarks primarily assess verbal reasoning, neglecting the complexities of non-verbal, multi-step visual simulation. We introduce STARE (Spatial Transformations and Reasoning Evaluation), a benchmark designed to evaluate multimodal large language models on tasks better solved through multi-step visual simulation. STARE features ~4K tasks spanning foundational geometric transformations (2D and 3D), integrated spatial reasoning (cube net folding and tangram puzzles), and real-world spatial reasoning (perspective and temporal reasoning), reflecting practical cognitive challenges like object assembly, mechanical diagram interpretation, and everyday spatial navigation. Our evaluations show that models excel at reasoning over simpler 2D transformations, but perform close to random chance on more complex tasks like 3D cube net folding and tangram puzzles that require multi-step visual simulations. Humans achieve near-perfect accuracy but take considerable time (up to 28.0s) on complex tasks, reducing response time by 7.5 seconds on average with intermediate visual simulations. In contrast, models exhibit inconsistent performance gains from visual simulations, improving on most tasks but declining in specific cases like tangram puzzles (GPT-4o, o1) and cube net folding (Claude-3.5, Gemini-2.0 Flash), indicating that models cannot consistently leverage intermediate visual information. Even o3, a strong reasoning model, lags significantly behind human performance across tasks. By evaluating non-verbal visual reasoning beyond conventional text-based benchmarks, STARE highlights critical gaps in current AI spatial capabilities and sets a new standard for assessing spatial intelligence in multimodal models.

Efficient processing of high-resolution images is crucial for real-world vision–language applications. However, existing Large Vision-Language Models (LVLMs) incur substantial computational overhead due to the large number of vision tokens. With the advent of "thinking with images" models, reasoning now extends beyond text to the visual domain. This capability motivates our two-stage "coarse-to-fine" reasoning pipeline: first, a downsampled image is analyzed to identify task-relevant regions; then, only these regions are cropped at full resolution and processed in a subsequent reasoning stage. This approach reduces computational cost while preserving fine-grained visual details where necessary. A major challenge lies in inferring which regions are truly relevant to a given query. Recent related methods often fail in the first stage after input-image downsampling, due to perception-driven reasoning, where clear visual information is required for effective reasoning. To address this issue, we propose ERGO (Efficient Reasoning & Guided Observation) that performs reasoning-driven perception—leveraging multimodal context to determine where to focus. Our model can account for perceptual uncertainty, expanding the cropped region to cover visually ambiguous areas for answering questions. To this end, we develop simple yet effective reward components in a reinforcement learning framework for coarse-to-fine perception. Across multiple datasets, our approach delivers higher accuracy than the original model and competitive methods, with greater efficiency. For instance, ERGO surpasses Qwen2.5-VL-7B on the V* benchmark by 4.7 points while using only 23% of the vision tokens, achieving a 3× inference speedup. The code is available at https://github.com/nota-github/ERGO.

Temporally localizing user-queried events through natural language is a crucial capability for video models. Recent methods predominantly adapt video LLMs to generate event boundary timestamps for temporal localization tasks, which struggle to leverage LLMs' pre-trained semantic understanding capabilities due to the uninformative nature of timestamp outputs. In this work, we explore a timestamp-free, semantic-oriented framework that fine-tunes video LLMs using two generative learning tasks and one discriminative learning task. We first introduce a structural token generation task that enables the video LLM to recognize the temporal structure of input videos based on the input query. Through this task, the video LLM generates a sequence of special tokens, called structural tokens, which partition the video into consecutive segments and categorize them as either target events or background transitions. To enhance precise recognition of event segments, we further propose a query-focused captioning task that enables the video LLM to extract fine-grained event semantics that can be effectively utilized by the structural tokens. Finally, we introduce a structural token grounding module driven by contrastive learning to associate each structural token with its corresponding video segment, achieving holistic temporal segmentation of the input video and readily yielding the target event segments for localization. Extensive experiments across diverse temporal localization tasks demonstrate that our proposed framework, MeCo, consistently outperforms methods relying on boundary timestamp generation, highlighting the potential of a semantic-driven approach for temporal localization with video LLMs.

Visual geolocalization, the task of predicting where an image was taken, remains challenging due to global scale, visual ambiguity, and the inherently hierarchical structure of geography. Existing paradigms rely on either large-scale retrieval, which requires storing a large number of image embeddings, grid-based classifiers that ignore geographic continuity, or generative models that diffuse over space but struggle with fine detail. We introduce an entity-centric formulation of geolocation that replaces image-to-image retrieval with a compact hierarchy of geographic entities embedded in Hyperbolic space. Images are aligned directly to country, region, subregion, and city entities through Geo-Weighted Hyperbolic contrastive learning by directly incorporating haversine distance into the contrastive objective. This hierarchical design enables interpretable predictions and efficient inference with 240k entity embeddings instead of over 5 million image embeddings on the OSV5M benchmark, on which our method establishes a new state-of-the-art performance. Compared to the current methods in the literature, it reduces mean geodesic error by 19.5\%, while improving the fine-grained subregion accuracy by 43\%. These results demonstrate that geometry-aware hierarchical embeddings provide a scalable and conceptually new alternative for global image geolocation.

Web agents promise to automate complex browser tasks, but current methods remain brittle -- relying on step-by-step UI interactions and heavy LLM reasoning that break under dynamic layouts and long horizons. Humans, by contrast, exploit website-provided functionality through high-level operations like search, filter, and sort. We introduce WALT (Web Agents that Learn Tools), a framework that reverse-engineers latent website functionality into deterministic, callable tools. Rather than hypothesizing ad-hoc skills, WALT exposes robust implementations of automations already designed into websites, spanning discovery (search, filter, sort), communication (post, comment, upvote), and content management (create, edit, delete). Tools abstract away low-level execution: instead of reasoning about how to click and type, agents simply call search(query) or create(listing). This shifts the computational burden from fragile step-by-step reasoning to reliable tool invocation. On VisualWebArena and WebArena, WALT achieves state-of-the-art success rates (52.9% on VisualWebArena, 50.1% on WebArena) with fewer steps and less LLM-dependent reasoning. On Online-Mind2Web, a benchmark of 139 real-world websites, WALT autonomously discovers 252 tools and improves success rate by 20.5% over a tool-free baseline, establishing a robust and generalizable paradigm for browser automation. Code: https://github.com/SalesforceAIResearch/WALT

Recent advances in multimodal learning have achieved remarkable success across diverse vision–language tasks. However, such progress heavily relies on large-scale image–text datasets, making training costly and inefficient. Prior efforts in dataset filtering and pruning attempt to mitigate this issue, but still require relatively large subsets to maintain performance and fail under very small subsets. Dataset distillation offers a promising alternative, yet existing multimodal dataset distillation methods require full-dataset training and joint optimization of image pixels and text features, making them architecture-dependent and limiting cross-architecture generalization. To overcome this, we propose a learning-free dataset distillation framework that eliminates the need for large-scale training and optimization while enhancing generalization across architectures. Our method uses CLIP to extract aligned image–text embeddings, obtains prototypes, and employs an unCLIP decoder to synthesize images, enabling efficient and scalable multimodal dataset distillation. Extensive experiments demonstrate that our approach consistently outperforms optimization-based dataset distillation and subset selection methods, achieving state-of-the-art cross-architecture generalization.

Multimodal Large Language Models (MLLMs) hold promise for accelerating scientific discovery by interpreting complex experimental procedures. However, their true capabilities are poorly understood, as existing benchmarks neglect the fine-grained and long-horizon nature of authentic laboratory work, especially in wet-lab settings. To bridge this gap, we introduce ExpVid, the first benchmark designed to systematically evaluate MLLMs on scientific experiment videos. Curated from peer-reviewed video publications, ExpVid features a new three-level task hierarchy that mirrors the scientific process: (1) Fine-grained Perception of tools, materials, and actions; (2) Procedural Understanding of step order and completeness; and (3) Scientific Reasoning that connects the full experiment to its published conclusions. Our vision-centric annotation pipeline, combining automated generation with multi-disciplinary expert validation, ensures that tasks require visual grounding. We evaluate 20 leading MLLMs on ExpVid and find that while they excel at coarse-grained recognition, they struggle with disambiguating fine details, tracking state changes over time, and linking experimental procedures to scientific outcomes. Our results reveal a notable performance gap between proprietary and open-source models, particularly in high-order reasoning. ExpVid not only provides a diagnostic tool but also charts a roadmap for developing MLLMs capable of becoming trustworthy partners in scientific experimentation.

Remote photoplethysmography (rPPG) enables non-contact physiological measurement but remains highly susceptible to illumination changes, motion artifacts, and limited temporal modeling. Large Language Models (LLMs) excel at capturing long-range dependencies, offering a potential solution but struggle with the continuous, noise-sensitive nature of rPPG signals due to their text-centric design. To bridge this gap, we introduce PhysLLM, a collaborative optimization framework that synergizes LLMs with domain-specific rPPG components. Specifically, the Text Prototype Guidance (TPG) strategy is proposed to establish cross-modal alignment by projecting hemodynamic features into LLM-interpretable semantic space, effectively bridging the representational gap between physiological signals and linguistic tokens. Besides, a novel Dual-Domain Stationary (DDS) Algorithm is proposed for resolving signal instability through adaptive time-frequency domain feature re-weighting. Finally, rPPG task-specific cues systematically inject physiological priors through physiological statistics, environmental contextual answering, and task description, leveraging cross-modal learning to integrate both visual and textual information, enabling dynamic adaptation to challenging scenarios like variable illumination and subject movements. Evaluation on four benchmark datasets, PhysLLM achieves state-of-the-art accuracy and robustness, demonstrating superior generalization across lighting variations and motion scenarios.

Large Vision-Language Models (LVLMs) have demonstrated remarkable multimodal capabilities, but they inherit the tendency to hallucinate from their underlying language models. While visual contrastive decoding has been proposed to mitigate this issue, existing methods often apply generic visual augmentations that disregard the specific context provided by the text query, limiting their effectiveness. This study introduces a novel training-free decoding strategy that addresses these limitations, featuring two key contributions. First, a self-augmentation prompting strategy that leverages the intrinsic knowledge of the model to dynamically align semantics between the query and the visual augmentation. Second, an adaptive thresholding algorithm that adaptively adjusts next token candidate size based on the output sparsity, utilizing full information from the logit distribution. Extensive experiments across four LVLMs and seven benchmarks demonstrate that the proposed decoding significantly enhances factual consistency compared to state-of-the-art decoding methods. This work highlights the importance of integrating query-dependent augmentation and entropy-aware decoding for improving effective generation of LVLMs. The source code will be released upon acceptance.

Industrial anomaly detection has been significantly advanced by Large Multimodal Models (LMMs), enabling diverse human instructions beyond detection, particularly through visually grounded reasoning for better image understanding. However, LMMs lack domain-specific knowledge, which limits their ability to generate accurate responses in complex industrial scenarios. In this work, we present JUDO, Juxtaposed Domain-Oriented Multimodal Reasoner, a framework that efficiently incorporates domain knowledge and context in visual and textual reasoning. Through visual reasoning, our model segments the defect region by juxtaposing query images with normal images as visual domain context, enabling a fine-grained visual comparative inspection. Furthermore, we inject domain knowledge through supervised fine-tuning (SFT) to enhance context understanding and subsequently guide domain reasoning through reinforcement learning (GRPO) with tailored rewards, opting for a domain-oriented reasoning process. Experimental results demonstrate that JUDO achieves superior performance on the MMAD benchmark, surpassing models such as Qwen2.5-VL-7B and GPT-4o. These results highlight the importance of enhancing domain knowledge and context for effective reasoning in anomaly understanding.

We introduce Part-X-MLLM, a native 3D multimodal large language model that unifies diverse 3D tasks by formulating them as programs in a structured, executable grammar. Given an RGB point cloud and a natural language prompt, our model autoregressively generates a single, coherent token sequence encoding part-level bounding boxes, semantic descriptions, and edit commands. This structured output serves as a versatile interface to drive downstream geometry-aware modules for part-based generation and editing. By decoupling the symbolic planning from the geometric synthesis, our approach allows any compatible geometry engine to be controlled through a single, language-native frontend. We pre-train a dual-encoder architecture to disentangle structure from semantics and instruction-tune the model on a large-scale, part-centric dataset. Experiments demonstrate that our model excels at producing high-quality, structured plans, enabling state-of-the-art performance in grounded Q&A, compositional generation, and localized editing through one unified interface. Project page: https://chunshi.wang/Part-X-MLLM/

Deep reasoning is fundamental for solving complex tasks, especially in vision-centric scenarios that demand sequential, multimodal understanding. However, existing benchmarks typically evaluate agents with fully synthetic, single-turn queries, limited visual modalities, and lack a framework to assess reasoning quality over multiple steps as required in real-world settings. To address this, we introduce Agent-X, a large-scale benchmark for evaluating vision-centric agents’ multistep and deep reasoning capabilities in real-world, multimodal settings. AgentX features 828 agentic tasks with authentic visual contexts, including images, multi-image comparisons, videos, and instructional text. These tasks span six major agentic environments: general visual reasoning, web browsing, security and surveillance, autonomous driving, sports, and math reasoning. Our benchmark requires agents to integrate tool use with explicit, stepwise decision-making in these diverse settings. In addition, we propose a fine-grained, step-level evaluation framework that assesses the correctness and logical coherence of each reasoning step and the effectiveness of tool usage throughout the task. Our results reveal that even the best-performing models, including GPT, Gemini, and Qwen families, struggle to solve multi-step vision tasks, achieving less than 50% full-chain success. These findings highlight key bottlenecks in current LMM reasoning and tool-use capabilities and identify future research directions in vision-centric agentic reasoning models

The rapid rise of image generation calls for detection methods that are both interpretable and reliable. Existing approaches, though accurate, act as black boxes and fail to generalize to out-of-distribution data, while multi-modal large language models (MLLMs) provide reasoning ability but often hallucinate. To address these issues, we construct \textbf{FakeXplained} dataset of AI-generated images annotated with bounding boxes and descriptive captions that highlight synthesis artifacts, forming the basis for human-aligned, visually grounded reasoning. Leveraging \textbf{FakeXplained}, we develop \textbf{FakeXplainer} which fine-tunes MLLMs with a progressive training pipeline, enabling accurate detection, artifact localization, and coherent textual explanations. Extensive experiments show that \textbf{FakeXplainer} not only sets a new state-of-the-art in detection and localization accuracy ($98.2\%$ accuracy, $36.0\%$ IoU), but also demonstrates strong robustness and out-of-distribution generalization, uniquely delivering spatially grounded, human-aligned rationales. The code and dataset are available at: \href{https://github.com/Gennadiyev/FakeXplain}{https://github.com/Gennadiyev/FakeXplain}.

Knowledge-Based Visual Question Answering (KB-VQA) requires models to answer questions about an image by integrating external knowledge, posing significant challenges due to noisy retrieval and the structured, encyclopedic nature of the knowledge base. These characteristics create a distributional gap from pretrained multimodal large language models (MLLMs), making effective reasoning and domain adaptation difficult in the post-training stage. In this work, we propose \textit{Wiki-R1}, a data-generation-based curriculum reinforcement learning framework that systematically incentivizes reasoning in MLLMs for KB-VQA. Wiki-R1 constructs a sequence of training distributions aligned with the model’s evolving capability, bridging the gap from pretraining to the KB-VQA target distribution. We introduce \textit{controllable curriculum data generation}, which manipulates the retriever to produce samples at desired difficulty levels, and a \textit{curriculum sampling strategy} that selects informative samples likely to yield non-zero advantages during RL updates. Sample difficulty is estimated using observed rewards and propagated to unobserved samples to guide learning. Experiments on two KB-VQA benchmarks, Encyclopedic VQA and InfoSeek, demonstrate that Wiki-R1 achieves new state-of-the-art results, improving accuracy from 35.5\% to 37.1\% on Encyclopedic VQA and from 40.1\% to 44.1\% on InfoSeek. The project page is available at https://artanic30.github.io/project_pages/WikiR1/.

Generating interactive 3D scenes from text requires not only synthesizing assets but arranging them with spatial intelligence—support, affordances, and plausibility. However, training data for interactive scenes is dominated by a few indoor datasets, so learning-based methods overfit to in-distribution layouts and struggle to compose diverse arrangements (e.g., outdoor settings and small-on-large relations). Meanwhile, LLM-based layout planners can propose diverse arrangements, but the lack of visual grounding often yields implausible placements that violate commonsense physics. We propose Scenethesis, a training-free, agentic framework that couples LLM-based scene planning with vision-guided layout refinement. Given a text prompt, Scenethesis first drafts a coarse layout with an LLM; a vision module refines the layout and extracts scene structure to capture inter-object relations. A novel optimization stage enforces pose alignment and physical plausibility, and a final judge verifies spatial coherence and triggers targeted repair when needed. Across indoor and outdoor prompts, Scenethesis produces realistic, relation-rich, and physically plausible 3D interactive scenes, reducing collisions and stability failures compared to SOTA methods, making it practical for virtual content creation, simulation, and embodied AI.

Understanding how humans interact with the surrounding environment, and specifically reasoning about object interactions and affordances, is a critical challenge in computer vision, robotics, and AI. Current approaches often depend on labor-intensive, hand-labeled datasets capturing real-world or simulated human-object interaction (HOI) tasks, which are costly and time-consuming to produce. Furthermore, most existing methods for 3D affordance understanding are limited to contact-based analysis, neglecting other essential aspects of human-object interactions, such as orientation (e.g., humans might have a preferential orientation with respect certain objects, such as a TV) and spatial occupancy (e.g., humans are more likely to occupy certain regions around an object, like the front of a microwave rather than its back). To address these limitations, we introduce H2OFlow, a novel framework that comprehensively learns 3D HOI affordances ---encompassing contact, orientation, and spatial occupancy--- using only synthetic data generated from 3D generative models. H2OFlow employs a dense 3D-flow-based representation, learned through a dense diffusion process operating on point clouds. This learned flow enables the discovery of rich 3D affordances without the need for human annotations. Through extensive quantitative and qualitative evaluations, we demonstrate that H2OFlow generalizes effectively to real-world objects and surpasses prior methods that rely on manual annotations or mesh-based representations in modeling 3D affordance.

Humans can imagine and manipulate visual images mentally, a capability known as \textit{spatial visualization}. While many multi-modal benchmarks assess reasoning on visible visual information, the ability to infer unseen relationships through spatial visualization remains insufficiently evaluated as a spatial skill. This reliance on publicly sourced problems from IQ tests or math competitions risks data contamination and compromises assessment reliability. To this end, we introduce \textbf{\textit{SpatialViz-Bench}}, a comprehensive multi-modal benchmark for \textit{spatial visualization} with \emph{12} tasks across \emph{4} sub-abilities, comprising \emph{1,180} programmatically generated problems, a scalable framework that allows for expansion to ensure fair and continuously reliable evaluations. Our evaluation of \emph{27} Multi-modal Large Language Models (MLLMs) reveals wide performance variations, demonstrates the benchmark's strong discriminative power, and uncovers counter-intuitive findings: Chain-of-Thought (CoT) prompting paradoxically degrades accuracy on open-source models. Through statistical and qualitative analysis of error types, SpatialViz-Bench demonstrates that state-of-the-art MLLMs exhibit deficiencies in \textit{spatial visualization} tasks, thereby addressing a significant lacuna in the field.

We introduce FLARE, a family of vision language models (VLMs) with a fully vision-language alignment and integration paradigm. Unlike existing approaches that rely on single MLP projectors for modality alignment and defer cross-modal interaction to LLM decoding, FLARE achieves deep, dynamic integration throughout the pipeline. Our key contributions include: (1) Text-Guided Vision Encoding that incorporates textual information during vision encoding to achieve pixel-level alignment; (2) Context-Aware Alignment Decoding that aggregates visual features conditioned on textual context during decoding for query-level integration; (3) Dual-Semantic Mapping Loss to supervise feature mapping from both modalities and enable modality-level bridging; and (4) Text-Driven VQA Synthesis that leverages high-quality text to generate VQA pairs and synthesize corresponding images, enabling data-level optimization. We train FLARE at 3B and 8B scales under both fixed and dynamic resolution settings, demonstrating that our full-modality alignment significantly outperforms existing methods while maintaining strong generalizability. FLARE 3B surpasses Cambrian-1 8B and Florence-VL 8B using only 630 vision tokens. Ablation studies reveal that FLARE achieves superior performance over existing methods with minimal computational cost. Even without dynamic resolution, FLARE outperforms LLaVA-NeXT, validating the effectiveness of our approach.

Spatial reasoning remains a fundamental challenge for Vision-Language Models (VLMs), with current approaches struggling to achieve robust performance despite recent advances. We identify that this limitation stems from a critical gap: existing methods attempt to learn spatial reasoning directly without establishing the hierarchical foundations of perception and understanding. To address this challenge, we present a comprehensive methodology for building spatial intelligence progressively. We introduce SpatialLadder-26k, a multimodal dataset containing 26,610 samples spanning object localization, single-image, multi-view, and video spatial reasoning tasks, constructed through a standardized pipeline that ensures systematic coverage across modalities. Building on this dataset, we design a three-stage progressive training framework that (1) establishes spatial perception through object localization, (2) develops spatial understanding through multi-dimensional spatial tasks, and (3) strengthens complex reasoning via reinforcement learning with verifiable rewards. This approach yields SpatialLadder, a 3B-parameter model that achieves state-of-the-art performance on spatial reasoning benchmarks, with 23.4% average improvement over the base model, surpassing GPT-4o by 20.8% and Gemini-2.0-Flash by 10.1%. Notably, SpatialLadder maintains strong generalization with 7.2% improvement on out-of-domain benchmarks, demonstrating that progressive training from perception to reasoning is essential for robust spatial intelligence.

Fully open multimodal large language models (MLLMs) currently lag behind proprietary counterparts, primarily due to a significant gap in data quality for supervised fine-tuning (SFT). Existing open-source datasets are often plagued by widespread noise and a critical deficit in complex reasoning data, such as Chain-of-Thought (CoT), which hinders the development of advanced model capabilities. Addressing these challenges, our work makes three primary contributions. First, we introduce Honey-Data-15M, a new SFT dataset comprising approximately 15 million QA pairs, processed through multiple cleaning techniques and enhanced with a novel dual-level (short and long) CoT enrichment strategy. Second, we introduce HoneyPipe, the data curation pipeline, and its underlying framework DataStudio, providing the community with a transparent and adaptable methodology for data curation that moves beyond static dataset releases. Finally, to validate our dataset and pipeline, we train Bee-8B, an 8B model on Honey-Data-15M. Experiments show that Bee-8B establishes a new state-of-the-art (SOTA) for fully open MLLMs, achieving performance that is competitive with, and in some cases surpasses, recent semi-open models such as InternVL3.5-8B. A comprehensive ablation study further dissects the impact of our data curation process, revealing that each stage provides significant performance gains across a wide range of benchmarks. Our work delivers to the community a suite of foundational resources, including: the Honey-Data-15M corpus; the full-stack suite comprising HoneyPipe and DataStudio; training recipes; an evaluation harness; and the model weights. This effort demonstrates that a principled focus on data quality is a key pathway to developing fully open MLLMs that are highly competitive with their semi-open counterparts.

Despite advancements in Multimodal Large Language Models (MLLMs), their proficiency in fine-grained video motion understanding remains critically limited. They often lack inter-frame differencing and tend to average or ignore subtle visual cues. Furthermore, while visual prompting has shown potential in static images, its application to videos' temporal complexities, particularly for fine-grained motion understanding, remains largely unexplored. We investigate whether inherent capability can be unlocked to boost MLLMs' motion perception and enable distinct visual signatures tailored to decouple object and camera motion cues. In this study, we introduce $\mathtt{MotionSight}$, a novel zero-shot method pioneering object-centric visual spotlight and motion blur as visual prompts to effectively improve fine-grained motion understanding without training. To convert this into valuable data assets, we curated $\mathtt{MotionVid-QA}$, the first large-scale dataset for fine-grained video motion understanding, with hierarchical annotations including SFT and preference data, $\Theta{(40K)}$ video clips and $\Theta{(87K)}$ QAs. Experiments show $\mathtt{MotionSight}$ achieves state-of-the-art open-source performance and competitiveness with commercial models. Using $\mathtt{MotionVid-QA}$, we fine-tuned $\mathtt{MotionChat}$ on Qwen2.5VL-7B, which attains 48.3\% overall accuracy on FAVOR-Bench that is comparable to Qwen2.5VL-72B's 48.1\%. In summary, we present a novel zero-shot method and a large-scale, high-quality dataset specifically for fine-grained motion understanding. All the code and annotations will be publicly available.

The aspiration for artificial general intelligence, fueled by the rapid progress of multimodal understanding, demands models to understand humans in diverse and complex scenarios, as humans manifests intelligence and embody the world. We propose HumanPCR, an evaluation suite for probing MLLMs’ capacity in human-centric visual contexts across three hierarchical levels: Perception, Comprehension, and Reasoning (denoted by Human-P, Human-C, and Human-R, respectively). Human-P and Human-C consist of over 6,000 multiple-choice questions evaluating 34 fine-grained tasks covering 9 essential dimensions. Human-R presents a manually curated challenging video reasoning test that requires integrating multiple visual evidence, proactively extracting implicit context beyond question cues, and applying human-like expertise. Each question includes human-annotated Chain-of-Thought (CoT) rationales with key visual evidence to support further research. Extensive evaluations on over 30 state-of-the-art models exhibit significant challenges in human-centric visual understanding, particularly in tasks involving detailed space perception, temporal understanding, and mind modeling. The analysis of Human-R further exposes a critical failure in reasoning: models struggle to proactively gather necessary visual evidence, instead showing a faulty reliance on query-prompted cues, with advanced techniques offering only marginal gains. We hope HumanPCR and our findings will advance the development, evaluation, and human-centric applications of multimodal models.

Direct Preference Optimization (DPO) has recently been extended from text-only models to vision-language models. However, existing methods rely on oversimplified pairwise comparisons, generating a single negative image via basic perturbations or similarity-based retrieval, which fail to capture the complex nature of multimodal preferences, inducing optimization bias and hallucinations. To address this issue, we propose MISP-DPO, the first framework to incorporate \emph{multiple}, semantically \emph{diverse} negative images in multimodal DPO via the Plackett-Luce model. Our method embeds prompts and candidate images in CLIP (Contrastive Language–Image Pre-training) space and applies a sparse autoencoder to uncover semantic deviations into interpretable factors. Negative samples are selected based on reconstruction difficulty, semantic deviation from the positive, and mutual diversity, yielding broader and more informative supervision. To handle multi-negative comparisons, we adopt a Plackett–Luce objective and introduce an importance sampling strategy that improves training efficiency. Experiments across five diverse benchmarks demonstrate that MISP-DPO consistently improves multimodal alignment over prior methods, validating the effectiveness of semantic-aware, multi-negative sampling in preference-based learning.

Despite recent advancements in extending context windows of large language models (LLMs) and large vision-language models (LVLMs), their ability to perform complex multi-modal reasoning over extended contexts remains critically limited. To underline this challenge, we present \textbf{MMReD}, a benchmark specifically designed to assess reasoning abilities within dense, information-rich scenarios where simple retrieval is not enough. Unlike traditional Needle-in-a-Haystack evaluations, MMReD challenges models to identify and interpret global patterns across entire contexts. Our benchmark comprises 24 tasks of varying complexity, ranging from standard passkey retrieval setups to those requiring selective or uniform attention to all context chunks. The evaluation reveals a consistent performance drop across all tested models -- including the most advanced LLMs, LVLMs, and architectures specializing in code and reasoning -- as the number of observations increases. Notably, even the leading reasoning-specialized models achieve 0\% accuracy on certain tasks at the maximum context length of 128 observations. Conventional fine-tuning techniques, such as SFT and GRPO, also fail to generalize effectively to longer contexts. These observations reveal an inherent limitation in current model architectures, emphasizing the need for innovative approaches to enable competent dense context reasoning in multi-modal AI systems.

Vision-Language Models (VLMs) often suffer from visual hallucinations -- generating things that are not consistent with visual inputs -- and language shortcuts, where they skip the visual part and just rely on text priors. These issues arise because most post-training methods for VLMs rely on simple verifiable answer matching and supervise only final outputs, leaving intermediate visual reasoning without explicit guidance. As a result, VLMs receive sparse visual signals and often learn to prioritize language-based reasoning over visual perception. To mitigate this, some existing methods add visual supervision using human annotations or distilled labels from external large models. However, human annotations are labor-intensive and costly, and external signals can introduce high latency cost. In this paper, we introduce Vision-SR1, a three-stage self-rewarding reinforcement learning method that improves visual reasoning without relying on external visual supervision. Vision-SR1 decomposes VLM reasoning into two components: visual reasoning and language reasoning, where the model is first prompted to produce self-contained visual descriptions sufficient to answer the question without referring back to the input image, before jointly optimizing both visual and language reasoning through our multi-reward loss objective. To validate this self-containment, the same VLM model is re-prompted to perform language reasoning using only the generated visual reasoning as input to compute visual reward. The final reward is computed through a decoupled reward-advantage framework, where visual reward and language reasoning reward each have their advantages, log probabilities, and KL divergence calculated separately. This decoupling enables more fine-grained reward computation by preventing the entanglement of heterogeneous reward signals. Our experiments show that Vision-SR1 improves visual reasoning, mitigates visual hallucinations, and reduces reliance on language shortcuts across diverse vision-language tasks, while being more efficient than methods that rely on external visual reward models, which require additional GPUs to host. In contrast, Vision-SR1 introduces no extra GPU overhead beyond that of standard training.

Multimodal Large Language Models (MLLMs) have achieved remarkable progress in vision-language tasks yet remain limited in long video understanding due to the limited context window. Consequently, prevailing approaches tend to rely on uniform frame sampling or static pre-selection, which might overlook critical evidence and unable to correct its initial selection error during its reasoning process. To overcome these limitations, we propose VideoZoomer, a novel agentic framework that enables MLLMs to dynamically control their visual focus during reasoning. Starting from a coarse low-frame-rate overview, VideoZoomer invokes a temporal zoom tool to obtain high-frame-rate clips at autonomously chosen moments, thereby progressively gathering fine-grained evidence in a multi-turn interactive manner. Accordingly, we adopt a two-stage training strategy: a cold-start supervised fine-tuning phase on a curated dataset of distilled exemplar and reflection trajectories, followed by reinforcement learning to further refine the agentic policy. Extensive experiments demonstrate that our 7B model demonstrates diverse and complex reasoning patterns, yielding strong results across a broad set of long video understanding and reasoning benchmarks. These emergent capabilities allow it to consistently surpass existing open-source models and even rival proprietary systems on challenging tasks, while achieving superior efficiency under reduced frame budgets.

Web agents such as deep research have demonstrated superhuman cognitive abilities, capable of solving highly challenging information-seeking problems. However, most research remains largely text-centric, overlooking visual information in the real world. This makes multimodal deep research highly challenging, as such agents require much stronger perceptual, logical, and knowledge-based reasoning abilities, as well as proficiency in more sophisticated tools. To address this limitation, we introduce WebWatcher, a multimodal agent for deep research with joint reasoning ability across both visual and textual modalities. It uses high-quality synthetic trajectories for efficient cold start training, utilizes various tools for deep reasoning, and further enhances generalization through reinforcement learning. To better evaluate the capabilities of multimodal agents, we propose BrowseComp-VL, a benchmark with the style of BrowseComp that requires complex information retrieval involving both visual and textual information. Experimental results show that WebWatcher outperforms the prompt-based workflow and open-source agents on HLE and BrowseComp-VL, and demonstrates its perception, multimodal reasoning, and searching capabilities across the other three benchmarks, respectively.

Online services rely on CAPTCHAs as a first line of defense against automated abuse, yet recent advances in multi-modal large language models (MLLMs) have eroded the effectiveness of conventional designs that focus on text recognition or 2D image understanding. To address this challenge, we present Spatial CAPTCHA, a novel human-verification framework that leverages fundamental differences in spatial reasoning between humans and MLLMs. Unlike existing CAPTCHAs that rely on low-level perception tasks vulnerable to modern AI, Spatial CAPTCHA generates dynamic questions requiring geometric reasoning, perspective-taking, occlusion handling, and mental rotation—skills intuitive for humans but difficult for current AI systems. The system employs a procedural generation pipeline with constraint-based difficulty control, automated correctness verification, and human-in-the-loop validation to ensure scalability, robustness, and adaptability. Evaluation on a corresponding benchmark, Spatial-CAPTCHA-Bench, demonstrates that humans vastly outperform 10 state-of-the-art MLLMs, with the best model achieving only 31.0\% Pass@1 accuracy. Result comparison with Google reCAPTCHA further confirms the effectiveness of Spatial CAPTCHA as both a security mechanism and a diagnostic tool for spatial reasoning in AI.

Visual Question Answering (VQA) is a core task for evaluating the capabilities of Vision–Language Models (VLMs). Existing VQA benchmarks primarily feature clear and unambiguous image–question pairs, whereas real-world scenarios often involve varying degrees of ambiguity that require nuanced reasoning and context-appropriate response strategies. Although recent studies have begun to address ambiguity in VQA, they lack (1) a systematic categorization of ambiguity levels and (2) datasets and models that support strategy-aware responses. In this paper, we introduce Ambiguous Visual Question Answering (AQuA), a fine-grained dataset that classifies ambiguous VQA instances into four levels according to the nature and degree of ambiguity, along with the optimal response strategy for each case. Our evaluation of diverse open-source and proprietary VLMs shows that most models fail to adapt their strategy to the ambiguity type, frequently producing overconfident answers rather than seeking clarification or acknowledging uncertainty. To address this challenge, we fine-tune VLMs on AQuA, enabling them to adaptively choose among multiple response strategies, such as directly answering, inferring intent from contextual cues, listing plausible alternatives, or requesting clarification. VLMs trained on AQuA achieve strategic response generation for ambiguous VQA, demonstrating the ability to recognize ambiguity, manage uncertainty, and respond with context-appropriate strategies, while outperforming both open-source and closed-source baselines.

Multimodal large language models (MLLMs) demonstrate strong video understanding by attending to visual tokens relevant to textual queries. To directly adapt this for localization in a training-free manner, we cast video reasoning segmentation as a video QA task and extract attention maps via rollout mechanism. However, raw attention maps are noisy and poorly aligned with object regions. We propose Decomposed Attention Fusion (DecAF), which refines these maps through two mechanisms: (1) contrastive object-background fusion and (2) complementary video-frame fusion. This method suppresses irrelevant activations and enhances object-focused cues, enabling direct conversion of attention maps into coarse segmentation masks. In addition, we introduce attention-guided SAM2 prompting for obtaining fine-grained masks. Unlike existing methods that jointly train MLLMs with SAM, our method operates entirely without retraining. DecAF outperforms training-free methods and achieves performance comparable to training-based methods on both referring and reasoning VOS benchmarks.

While table understanding increasingly relies on pixel-only settings, current benchmarks predominantly use synthetic renderings that lack the complexity and visual diversity of real-world tables. Additionally, existing visual table understanding (VTU) datasets offer fixed examples with single visualizations and pre-defined instructions, providing no access to underlying serialized data for reformulation. We introduce TABLET, a large-scale VTU dataset with 4 million examples across 21 tasks, grounded in 2 million unique tables where 88% preserve original visualizations. To evaluate whether models are able to jointly reason over tabular and visual content, we also introduce VisualTableQA, a benchmark requiring both visual perception and table understanding. Fine-tuning vision-language models like Qwen2.5-VL-7B and Gemma 3-4B on TABLET improves performance on seen and unseen VTU tasks while increasing robustness on real-world table visualizations. By preserving original visualizations and maintaining example traceability in a unified large-scale collection, TABLET establishes a foundation for robust training and extensible evaluation of future VTU models.

Recent advances in 3D Large Language Models (3D-LLMs) show strong potential in understanding and interacting with 3D environments, yet their training data typically lack explicit reasoning processes, limiting complex spatial reasoning and task planning. To address this, we annotate SCoT, a million-scale Chain-of-Thought dataset spanning three levels: a) Spatial Perception (what is there), recognizing object properties, relations, and scene attributes; b) Spatial Analysis (what does it mean), inferring rationality, functionalities, and physical implications; c) Spatial Planning (what should I do), integrating perception and reasoning for actionable strategies. Unlike prior datasets supervising only answers, SCoT annotates intermediate reasoning grounded in scene cues, specifically for analysis and planning tasks. Results show that CoT supervision greatly benefits complex analysis and planning but induces hallucinations and accuracy drops in simple perception. These findings highlight both the necessity and the nuanced challenges of scene-grounded reasoning for advancing 3D intelligence.

Multimodal large language models (MLLMs) suffer from high computational costs due to excessive visual tokens, particularly in high-resolution and video-based scenarios. Existing token reduction methods typically focus on isolated pipeline components and often neglect textual alignment, leading to performance degradation. In this paper, we propose VisionTrim, a unified framework for training-free MLLM acceleration, integrating two effective plug-and-play modules: 1) the Dominant Vision Token Selection (DVTS) module, which preserves essential visual tokens via global-local view, and 2) the Text-Guided Vision Complement (TGVC) module, which facilitates context-aware token merging guided by textual cues. Extensive experiments across diverse image and video multimodal benchmarks demonstrate the performance superiority of our VisionTrim, advancing practical MLLM deployment in real-world applications. The code is available at: https://github.com/hanxunyu/VisionTrim.

Videos, with their unique temporal dimension, demand precise grounded understanding, where answers are directly linked to visual, interpretable evidence. Despite significant breakthroughs in text-based reasoning with large language models, multi-modal reasoning - especially for videos - remains limited. In this work, we fill this gap by introducing VideoMind, a novel video-language agent for temporal-grounded video reasoning. Our method involves two key innovations: (1) We identify four essential capabilities for grounded video reasoning and propose a role-based agentic workflow, comprising a planner to coordinate roles, a grounder for temporal event localization, a verifier to assess event candidates, and an answerer for question answering. (2) To efficiently integrate these roles during inference, we propose a novel Chain-of-LoRA mechanism, where a unified base model with multiple LoRA adapters is leveraged to enable seamless role switching, balancing efficiency and flexibility. Extensive experiments on 15 benchmarks across Grounded VideoQA, Video Temporal Grounding, and General VideoQA tasks demonstrate the effectiveness of the proposed scheme in advancing video agent, test-time scaling, and long-form video reasoning. Code, models, datasets, and demos are available at https://videomind.github.io/.

Large Vision-Language Models (LVLMs) achieve impressive performance across multiple tasks. A significant challenge, however, is their prohibitive inference cost when processing high-resolution visual inputs. While visual token pruning has emerged as a promising solution, existing methods that primarily focus on semantic relevance often discard tokens that are crucial for spatial reasoning. We address this gap through a novel insight into how LVLMs process spatial reasoning. Specifically, we reveal that LVLMs implicitly establish visual coordinate systems through Rotary Position Embeddings (RoPE), where specific token positions serve as implicit visual coordinates (IVC tokens) that are essential for spatial reasoning. Based on this insight, we propose IVC-Prune, a training-free, prompt-aware pruning strategy that retains both IVC tokens and semantically relevant foreground tokens. IVC tokens are identified by theoretically analyzing the mathematical properties of RoPE, targeting positions at which its rotation matrices approximate identity matrix or the $90^\circ$ rotation matrix. Foreground tokens are identified through a robust two-stage process: semantic seed discovery followed by contextual refinement via value-vector similarity. Extensive evaluations across four representative LVLMs and twenty diverse benchmarks show that IVC-Prune reduces visual tokens by approximately 50\% while maintaining $\geq$ 99\% of the original performance and even achieving improvements on several benchmarks.

Bongard Problems (BPs) provide a challenging testbed for abstract visual reasoning (AVR), requiring models to identify visual concepts from just a few examples and describe them in natural language. Early BP benchmarks featured synthetic black-and-white drawings, which might not fully capture the complexity of real-world scenes. Subsequent BP datasets employed real-world images, albeit the represented concepts are identifiable from high-level image features, reducing the task complexity. Differently, the recently released Bongard-RWR dataset aimed at representing abstract concepts formulated in the original BPs using fine-grained real-world images. Its manual construction, however, limited the dataset size to just $60$ instances, constraining evaluation robustness. In this work, we introduce Bongard-RWR+, a BP dataset composed of $5400$ instances that represent original BP abstract concepts using real-world-like images generated via a vision language model (VLM) pipeline. Building on Bongard-RWR, we employ Pixtral-12B to describe manually curated images and generate new descriptions aligned with the underlying concepts, use Flux.1-dev to synthesize images from these descriptions, and manually verify that the generated images faithfully reflect the intended concepts. We evaluate state-of-the-art VLMs across diverse BP formulations, including binary and multiclass classification, as well as textual answer generation. Our findings reveal that while VLMs can recognize coarse-grained visual concepts, they consistently struggle with discerning fine-grained concepts, highlighting limitations in their reasoning capabilities.

Large multimodal models have achieved remarkable progress in integrating vision and language, enabling strong performance across perception, reasoning, and domain-specific tasks. However, their capacity to reason over multiple, visually similar inputs remains insufficiently explored. Such fine-grained comparative reasoning is central to real-world tasks, especially in mathematics and education, where learners must often distinguish between nearly identical diagrams to identify correct solutions. To address this gap, we present VisioMath, a curated benchmark of 1,800 high-quality K–12 mathematics problems in which all candidate answers are diagrams with subtle visual similarities. A comprehensive evaluation of state-of-the-art LMMs, covering both leading closed-source systems and widely adopted open-source models, reveals a consistent decline in accuracy as inter-image similarity increases. Analysis indicates that the dominant failure mode stems from image–text misalignment: rather than grounding reasoning in textual cues, models often resort to shallow positional heuristics, resulting in systematic errors. We further explore three alignment-oriented strategies, spanning training-free approaches and finetuning, and achieve substantial accuracy gains. We hope that VisioMath will serve as a rigorous benchmark and catalyst for developing LMMs toward deeper diagram understanding, precise comparative reasoning, and grounded multi-image–text integration. The code and dataset are available at https://github.com/Nefefilibata/VisioMath.

The webpage-to-code task requires models to understand visual representations of webpages and generate corresponding code. However, existing benchmarks primarily focus on static screenshot-to-code tasks, thereby overlooking the dynamic interactions fundamental to real-world web applications. To address this limitation, this paper introduces IWR-Bench, a novel benchmark for evaluating the capabilities of Large Vision-Language Models (LVLMs) in interactive webpage reconstruction from video. IWR-Bench comprises 113 meticulously curated tasks from 100 real-world websites, with 1,001 actions and featuring diverse interaction complexities (e.g., web games), visual styles, and domains. Aligning with standard web development practices, each task includes not only user interaction videos but also all crawled static assets (e.g., images, videos). This benchmark evaluates models on two fundamental challenges: comprehensive multi-modal reasoning to infer interaction logic from video and assets, and advanced code generation to translate this logic into functional code. An agent-as-a-judge framework with a comprehensive metric system automatically assesses the functional correctness and visual fidelity of generated webpages. Extensive experiments on 28 LVLMs reveal a significant challenge: the best model achieves an overall score of only 36.35\%, as functional correctness (24.39\% IFS) lags significantly behind visual fidelity (64.25\% VFS). These results highlight critical limitations in current models' ability to reason about temporal dynamics and synthesize event-driven logic, establishing IWR-Bench as a challenging frontier for vision-language research. The benchmark and evaluation code will be made publicly available.

Reasoning Video Object Segmentation is a challenging task, aiming at generating a mask sequence from an input video given a complex and implicit text query. While existing works finetune Multimodal Large Language Models (MLLM) for the task, they still fail in video inputs given complex temporally-sensitive queries, indicating their lack of temporal and spatial integration in complex scenarios. In this paper, we propose CoT-RVS, a novel framework employing the zero-shot Chain-of-Thought (CoT) capability of MLLM to address these complex challenges by temporal-semantic reasoning: CoT-RVS analyzes the visible objects within a given frame that possibly match the language query (semantic), and chooses a corresponding keyframe for each object that can be observed effortlessly among all frames (temporal). Notably, the CoT-RVS framework is training-free and compatible with closed-source MLLMs, which can be applied to Reasoning Video Instance Segmentation. Our framework's training-free feature further allows its extension to process online video streams, where the CoT is used at test time to update the object of interest when a better target starts to emerge and becomes visible. We conduct extensive experiments on video object segmentation with explicit and implicit queries. The results show that CoT-RVS significantly outperforms previous works in both cases, qualitatively and quantitatively.

Multimodal large language models (MLLMs) represent images and video frames as visual tokens. Scaling from single images to hour-long videos, however, inflates the token budget far beyond practical limits. Popular pipelines therefore either uniformly subsample or apply keyframe selection with retrieval-style scoring using smaller vision-language models. However, these keyframe selection methods still rely on pre-filtering before selection to reduce the inference cost and can miss the most informative moments. We propose FOCUS, Frame-Optimistic Confidence Upper-bound Selection, a training-free, model-agnostic keyframe selection module that selects query-relevant frames under a strict token budget. FOCUS formulates keyframe selection as a combinatorial pure-exploration (CPE) problem in multi-armed bandits: it treats short temporal clips as arms, and uses empirical means and Bernstein confidence radius to identify informative regions while preserving exploration of uncertain areas. The resulting two-stage exploration-exploitation procedure reduces from a sequential policy with theoretical guarantees, first identifying high-value temporal regions, then selecting top-scoring frames within each region. Extensive experiments across four long-video question-answering benchmarks and four popular MLLMs demonstrate that FOCUS delivers substantial accuracy improvements while processing less than 2% of video frames. For videos longer than 20 minutes, it achieves an 11.9% gain in accuracy on LongVideoBench, demonstrating its effectiveness as a keyframe selection method and providing a simple and general solution for scalable long-video understanding with MLLMs.

In this paper, we propose a training scheme called OVSeg3R to learn open-vocabulary 3D instance segmentation from well-studied 2D perception models with the aid of 3D reconstruction. OVSeg3R directly adopts reconstructed scenes from 2D videos as input, avoiding costly manual adjustment while aligning input with real-world applications. By exploiting the 2D to 3D correspondences provided by 3D reconstruction models, OVSeg3R projects each view's 2D instance mask predictions, obtained from an open-vocabulary 2D model, onto 3D to generate annotations for the view's corresponding sub-scene. To avoid incorrectly introduced false positives as supervision due to partial annotations from 2D to 3D, we propose a View-wise Instance Partition algorithm, which partitions predictions to their respective views for supervision, stabilizing the training process. Furthermore, since 3D reconstruction models tend to over-smooth geometric details, clustering reconstructed points into representative super-points based solely on geometry, as commonly done in mainstream 3D segmentation methods, may overlook geometrically non-salient objects. We therefore introduce 2D Instance Boundary-aware Superpoint, which leverages 2D masks to constrain the superpoint clustering, preventing superpoints from violating instance boundaries. With these designs, OVSeg3R not only extends a state-of-the-art closed-vocabulary 3D instance segmentation model to open-vocabulary, but also substantially narrows the performance gap between tail and head classes, ultimately leading to an overall improvement of +2.3 mAP on the ScanNet200 benchmark. Furthermore, under the standard open-vocabulary setting, OVSeg3R surpasses previous methods by about +7.1 mAP on the novel classes, further validating its effectiveness.

Micro-expression recognition (MER) is highly challenging due to the subtle and rapid facial muscle movements and the scarcity of annotated data. Existing methods typically rely on pixel-level motion descriptors such as optical flow and frame difference, which tend to be sensitive to identity and lack generalization. In this work, we propose D-FACE, a Discrete Facial ACtion Encoding framework that leverages large-scale facial video data to pretrain an identity- and domain-invariant facial action tokenizer, for MER. For the first time, MER is shifted from relying on pixel-level motion descriptors to leveraging semantic-level facial action tokens, providing compact and generalizable representations of facial dynamics. Empirical analyses reveal that these tokens exhibit position-dependent semantics, motivating sequential modeling. Building on this insight, we employ a Transformer with sparse attention pooling to selectively capture discriminative action cues. Furthermore, to explicitly bridge action tokens with human-understandable emotions, we introduce an emotion-description-guided CLIP (EDCLIP) alignment. EDCLIP leverages textual prompts as semantic anchors for representation learning, while enforcing that the "others" category, which lacks corresponding prompts due to its ambiguity, remains distant from all anchor prompts. Extensive experiments on multiple datasets demonstrate that our method achieves not only state-of-the-art recognition accuracy but also high-quality cross-identity and even cross-domain micro-expression generation, suggesting a paradigm shift from pixel-level to generalizable semantic-level facial motion analysis. Code is available at https://github.com/KinopioIsAllIn/D-FACE.

Anomaly detection is a critical task in computer vision with profound implications for medical imaging, where identifying pathologies early can directly impact patient outcomes. While recent unsupervised anomaly detection approaches show promise, they require substantial normal training data and struggle to generalize across anatomical contexts. We introduce D$^2$4FAD, a novel dual distillation framework for few-shot anomaly detection that identifies anomalies in previously unseen tasks using only a small number of normal reference images. Our approach leverages a pre-trained encoder as a teacher network to extract multi-scale features from both support and query images, while a student decoder learns to distill knowledge from the teacher on query images and self-distill on support images. We further propose a learn-to-weight mechanism that dynamically assesses the reference value of each support image conditioned on the query, optimizing anomaly detection performance. To evaluate our method, we curate a comprehensive benchmark dataset comprising 13,084 images across four organs, four imaging modalities, and five disease categories. Extensive experiments demonstrate that D$^2$4FAD significantly outperforms existing approaches, establishing a new state-of-the-art in few-shot medical anomaly detection. Code is available at https://github.com/ttttqz/D24FAD.

This paper addresses the domain generalization (DG) problem in deep learning. While most DG methods focus on enforcing visual feature invariance, we leverage the reasoning capability of multimodal large language models (MLLMs) and explore the potential of constructing reasoning chains that derives image categories to achieve more robust predictions under domain shift. To this end, we systematically study the role of reasoning in DG using DomainBed-Reasoning, a newly constructed extension of DomainBed dataset, in which each sample is paired with class-relevant reasoning chains. Our analysis reveals two key challenges: (i) fine-tuning MLLMs with reasoning chains for classification is more challenging than direct label supervision, since the model must optimize complex reasoning sequences before label prediction; and (ii) mismatches in reasoning patterns between supervision signals and fine-tuned MLLMs lead to a trade-off between semantic richness (informative but harder to optimize) and optimization efficiency (easier to optimize but less informative). To address these issues, we propose RD-MLDG (Reasoning-Driven Multimodal LLM for Domain Generalization), a framework with two components: (i) MTCT (Multi-Task Cross-Training), which introduces an additional direct classification pathway to guide reasoning supervision; and (ii) SARR (Self-Aligned Reasoning Regularization), which preserves the semantic richness of reasoning chains while mitigating reasoning-pattern mismatches via iterative self-labeling. Experiments on standard DomainBed datasets (PACS, VLCS, OfficeHome, TerraIncognita) demonstrate that RD-MLDG achieves state-of-the-art performances, highlighting reasoning as a promising complementary signal for robust out-of-domain generalization.

Video recognition models are typically trained on fixed taxonomies which are often too coarse, collapsing distinctions in object, manner or outcome under a single label. As tasks and definitions evolve, such models cannot accommodate emerging distinctions and collecting new annotations and retraining to accommodate such changes is costly. To address these challenges, we introduce category splitting, a new task where an existing classifier is edited to refine a coarse category into finer subcategories, while preserving accuracy elsewhere. We propose a zero-shot editing method that leverages the latent compositional structure of video classifiers to expose fine-grained distinctions without additional data. We further show that low-shot fine-tuning, while simple, is highly effective and benefits from our zero-shot initialization. Experiments on our new video benchmarks for category splitting demonstrate that our method substantially outperforms vision-language baselines, improving accuracy on the newly split categories without sacrificing performance on the rest. Project page: https://kaitingliu.github.io/Category-Splitting/.

With the rise of deep neural networks, especially in safety-critical applications, robustness and interpretability are crucial to ensure their trustworthiness. Recent advances in 3D-aware classifiers that map image features to volumetric representation of objects, rather than relying solely on 2D appearance, have greatly improved robustness on out-of-distribution (OOD) data. Such classifiers have not yet been studied from the perspective of interpretability. Meanwhile, current concept-based XAI methods often neglect OOD robustness. We aim to address both aspects with CAVE - Concept Aware Volumes for Explanations - a new direction that unifies interpretability and robustness in image classification. We design CAVE as a robust and inherently interpretable classifier that learns sparse concepts from 3D object representation. We further propose 3D Consistency (3D-C), a metric to measure spatial consistency of concepts. Unlike existing metrics that rely on human-annotated parts on images, 3D-C leverages ground-truth object meshes as a common surface to project and compare explanations across concept-based methods. CAVE achieves competitive classification performance while discovering consistent and meaningful concepts across images in various OOD settings.

As fine-tuning becomes impractical at scale, probing is emerging as the preferred evaluation protocol. However, standard linear probing can understate the capability of models whose pre-training optimizes local representations rather than an explicit global representation. This motivates attentive probing, an alternative that uses attention to selectively aggregate patch-level features. Despite growing adoption, attentive probing is still underexplored: existing approaches are often over-parameterized and computationally inefficient. In this work, we revisit attentive probing through the lens of the accuracy vs. parameter-efficiency trade-off. We present the first comprehensive study of existing methods, analyzing their design choices and benchmarking their performance. Building on these insights, we propose efficient probing (EP), a lightweight yet effective multi-query cross-attention mechanism that eliminates redundant projections and reduces the number of trainable parameters. Across multiple benchmarks and pre-training paradigms, EP consistently outperforms linear probing and previous attentive probing methods, and remains effective when combined with parameter-efficient fine-tuning. Beyond evaluation, our analysis uncovers emerging properties of EP, including complementary attention maps, which open new directions for leveraging probing beyond protocol design. Project page: https://vrg.fel.cvut.cz/ep/.

The increasing demand for real-time interaction in online video scenarios necessitates a new class of efficient streaming video understanding models. However, existing approaches often rely on a query-agnostic ''change-is-important'' assumption, which conflates visual dynamics with semantic relevance, leading to computational redundancy and mistimed responses. To address this, we propose QueryStream, a novel framework that integrates query-awareness into the core of video processing and response scheduling. QueryStream features two synergistic components: (1) Query-Aware Differential Pruning (QDP), a policy that filters the token stream by jointly assessing semantic relevance to the query and temporal novelty against a dynamically smoothed history; and (2) Relevance-Triggered Active Response (RTAR), a dual-gated mechanism that schedules responses based on both high query relevance and significant information density. As a lightweight, training-free module, QueryStream achieves state-of-the-art performance on benchmarks such as StreamingBench and OVO-Bench under moderate pruning, and matches full-token baselines while pruning over 70\% of visual tokens. Notably, our pruning mechanism generalizes to offline tasks, where it serves as a context-denoising module that benefits long-form video understanding. This work not only reveals the vast semantic redundancy in video streams relative to user intent but also establishes a promising, intent-driven direction for efficient and robust online video understanding. Code is available at: https://github.com/Zhangkr2003/QueryStream.

The development of multimodal large language models (MLLMs) enables the evaluation of image quality through natural language descriptions. This advancement allows for more detailed assessments. However, these MLLM-based IQA methods primarily rely on general contextual descriptions, sometimes limiting fine-grained quality assessment. To address this limitation, we introduce a new image quality assessment (IQA) task paradigm, grounding-IQA. This paradigm integrates multimodal referring and grounding with IQA to realize more fine-grained quality perception, thereby extending existing IQA. Specifically, grounding-IQA comprises two subtasks: grounding-IQA-description (GIQA-DES) and visual question answering (GIQA-VQA). GIQA-DES involves detailed descriptions with precise locations (e.g., bounding boxes), while GIQA-VQA focuses on quality QA for local regions. To realize grounding-IQA, we construct a corresponding dataset, GIQA-160K, through our proposed automated annotation pipeline. Furthermore, we develop a well-designed benchmark, GIQA-Bench. The benchmark evaluates the grounding-IQA performance from three perspectives: description quality, VQA accuracy, and grounding precision. Experiments demonstrate that our proposed method facilitates the more fine-grained IQA application. Code: https://github.com/zhengchen1999/Grounding-IQA.

Traditional super-resolution (SR) methods assume an "ideal'' downscaling SR-kernel (e.g., bicubic downscaling) between the high-resolution (HR) image and the low-resolution (LR) image. Such methods fail once the LR images are generated differently. Current blind-SR methods aim to remove this assumption, but are still fundamentally restricted to rather simplistic downscaling SR-kernels (e.g., anisotropic Gaussian kernels), and fail on more complex (out of distribution) downscaling degradations. However, using the correct SR-kernel is often more important than using a sophisticated SR algorithm. In "KernelFusion'', we introduce a zero-shot diffusion-based method that uses an unrestricted kernel. Our method recovers the unique image-specific SR-kernel directly from the LR input image, while simultaneously recovering its corresponding HR image. KernelFusion exploits the principle that the correct SR-kernel is the one that maximizes patch similarity across different scales of the LR image. We first train an image-specific patch-based diffusion model on the single LR input image, capturing its unique internal patch statistics. We then reconstruct a larger HR image with the same learned patch distribution, while simultaneously recovering the correct downscaling SR-kernel that maintains this cross-scale relation between the HR and LR images. Empirical results demonstrate that KernelFusion handles complex downscaling degradations where existing Blind-SR methods fail, achieving robust kernel recovery and superior SR quality. By breaking free from predefined kernel assumptions and training distributions, KernelFusion establishes a new paradigm of zero-shot Blind-SR that can handle unrestricted, image-specific kernels previously thought impossible.

In this paper, we present DiffuDETR, a novel approach that formulates object detection as a conditional object query generation task, conditioned on the image and a set of noisy reference points. We integrate DETR-based models with denoising diffusion training to generate object queries' reference points from a prior gaussian distribution. We propose two variants: DiffuDETR, built on top of the Deformable DETR decoder, and DiffuDINO, based on DINO’s decoder with contrastive denoising queries (CDNs). To improve inference efficiency, we further introduce a lightweight sampling scheme that requires only multiple forward passes through the decoder. Our method demonstrates consistent improvements across multiple backbones and datasets, including COCO2017, LVIS, and V3Det, surpassing the performance of their respective baselines, with notable gains in complex and crowded scenes. Using ResNet-50 backbone we observe a +1.0 in COCO-val reaching 51.9 mAP on DiffuDINO compared to 50.9 mAP of the DINO. We also observe similar improvements on LVIS and V3DET datasets with +2.4 and +2.2 respectively. The code is available at https://github.com/MBadran2000/DiffuDETR.

Recently, considerable progress has been made in all-in-one image restoration. Generally, existing methods can be degradation-agnostic or degradation-aware. However, the former are limited in leveraging degradation estimation-based priors, and the latter suffer from the inevitable error in degradation estimation. Consequently, the performance of existing methods has a large gap compared to specific single-task models. In this work, we make a step forward in this topic, and present our UniRestorer with improved restoration performance. Specifically, we perform hierarchical clustering on degradation space, and train a multi-granularity mixture-of-experts (MoE) restoration model. Then, UniRestorer adopts both degradation and granularity estimation to adaptively select an appropriate expert for image restoration. In contrast to existing degradation-agnostic and -aware methods, UniRestorer can leverage degradation estimation to benefit degradation-specific restoration, and use granularity estimation to make the model robust to degradation estimation error. Experimental results show that our UniRestorer outperforms state-of-the-art all-in-one methods by a large margin, and is promising in closing the performance gap to specific single-task models. The code and pre-trained models will be publicly available.

The quadratic compute and memory costs of global self-attention severely limit its use in high-resolution images. Local attention reduces complexity by restricting attention to neighborhoods. Block-sparse kernels can further improve the efficiency of local attention, but conventional local attention patterns often fail to deliver significant speedups because tokens within a window are not contiguous in the 1D sequence. This work proposes a novel method for constructing windows and neighborhoods based on the Hilbert curve. Image tokens are first reordered along a Hilbert curve, and windows and neighborhoods are then formed on the reordered 1D sequence. From a block-sparse perspective, this strategy significantly increases block sparsity and can be combined with existing block-sparse kernels to improve the efficiency of 2D local attention. Experiments show that the proposed Hilbert Window Attention and Hilbert Slide Attention can accelerate window attention and slide attention by about $4\times$ and $18\times$, respectively. To assess practicality, the strategy is instantiated as the Hilbert Window Transformer and the Hilbert Neighborhood Transformer, both of which achieve end-to-end speedups with minimal accuracy loss. Overall, combining Hilbert-guided local attention with block-sparse kernels offers a general and practical approach to enhancing the efficiency of 2D local attention for images.

Within the context of representation learning for Earth observation, geographic Implicit Neural Representations (INRs) embed low-dimensional location inputs (longitude, latitude) into high-dimensional embeddings, through models trained on geo-referenced satellite, image or text data. Despite the common aim of geographic INRs to distill Earth's data into compact, learning-friendly representations, we lack an understanding of how much information is contained in these Earth representations, and where that information is concentrated. The intrinsic dimension of a dataset measures the number of degrees of freedom required to capture its local variability, regardless of the ambient high-dimensional space in which it is embedded. This work provides the first study of the intrinsic dimensionality of geographic INRs. Analyzing INRs with ambient dimension between 256 and 512, we find that their intrinsic dimensions fall roughly between 2 and 10 and are sensitive to changing spatial resolution and input modalities during INR pre-training. Furthermore, we show that the intrinsic dimension of a geographic INR correlates with downstream task performance and can capture spatial artifacts, facilitating model evaluation and diagnostics. More broadly, our work offers an architecture-agnostic, label-free metric of information content that can enable unsupervised evaluation, model selection, and pre-training design across INRs.

Emotion understanding is a critical yet challenging task. Recent advances in Multimodal Large Language Models (MLLMs) have significantly enhanced their capabilities in this area. However, MLLMs often suffer from ``hallucinations'', generating irrelevant or nonsensical content. To the best of our knowledge, and despite the importance of this issue, there has been no dedicated effort to evaluate emotion-related hallucinations in MLLMs. In this work, we introduce \textbf{EmotionHallucer}, the first benchmark for detecting and analyzing emotion hallucinations in MLLMs. Unlike humans, whose emotion understanding stems from the interplay of biology and social learning, MLLMs rely solely on data-driven learning and lack innate emotional instincts. Fortunately, emotion psychology provides a solid foundation of knowledge about human emotions. Building on this knowledge, we assess emotion hallucinations from two perspectives: emotion psychology knowledge and realworld multimodal perception. To support robust evaluation, we utilize an adversarial binary question–answer (QA) framework, which employs carefully crafted basic and hallucinated pairs to assess the emotion hallucination tendencies of MLLMs. By evaluating 41 LLMs and MLLMs on EmotionHallucer, we find that: (1) most current models exhibit substantial issues with emotion hallucinations; (2) closed-source models outperform open-source models in detecting emotion hallucinations, and reasoning capability provides additional advantages; and (3) existing models perform better in emotion psychology knowledge than in multimodal emotion perception. As a byproduct, these findings inspire us to propose the \textbf{PEP-MEK} framework, which yields an average improvement of 9.90\% in emotion hallucination detection across selected models. Resources will be available on GitHub.

Physical simulation relies on spatially-varying mechanical properties, typically laboriously hand-crafted. We present the first feed-forward model to predict fine-grained mechanical properties, Young’s modulus ($E$), Poisson’s ratio ($\nu$), and density ($\rho$), throughout *the volume* of 3D objects. Our model supports any 3D representation that can be rendered and voxelized, including Signed Distance Fields (SDFs), Gaussian Splats and Neural Radiance Fields (NeRFs). To achieve this, we aggregate per-voxel multi-view features for any input, which are passed to our trained Geometry Transformer to predict per-voxel material latent codes. These latents reside on the trained manifold of physically plausible materials, which we train on a real-world dataset, guaranteeing the validity of decoded per-voxel materials. To obtain object-level training data, we propose an annotation pipeline combining knowledge from segmented 3D datasets, material databases, and a vision-language model. Experiments show that VoMP estimates accurate volumetric properties and can convert 3D objects into simulation-ready assets, resulting in realistic deformable simulations and far outperforming prior art.

Dataset distillation (DD) attempts to construct a compact synthetic dataset that serves as a proxy for a large real dataset under a fixed storage budget, thereby reducing the storage burden and training costs. Prior works assume the full dataset is available upfront which is distilled at once, although real datasets are collected incrementally over time in practice. To alleviate this gap, we introduce a new problem setting, Domain Incremental Dataset Distillation, that continually distills datasets from different domains into a single synthetic dataset. The conventional DD sequentially processes arriving datasets in order, overwriting the old knowledge with new one, causing catastrophic forgetting problem. To overcome this drawback, we propose Asymmetric Synthetic Data Update strategy that adjusts the per-sample update rates for synthetic dataset while balancing the stability-plasticity trade-off. Specifically, we design a bi-level optimization method based on meta-learning framework to estimate the optimal update rates, which allows each sample to focus on either stability or plasticity, thereby striking a balance between them. Experimental results demonstrate that our approach effectively mitigates the catastrophic forgetting and achieves superior performance of DD across continually incoming datasets compared with existing methods.

This paper studies the problem of data-specific neural representations, aiming for compact, flexible, and modality-agnostic storage of individual visual data using neural networks. Our approach considers a visual datum as a set of discrete observations of an underlying continuous signal, thus requiring models capable of capturing the inherent structure of the signal. For this purpose, we investigate state-space models (SSMs), which are well-suited for modeling latent signal dynamics. We first explore the appealing properties of SSMs for data-specific neural representation and then present a novel framework that integrates SSMs into the representation pipeline. The proposed framework achieved compact representations and strong reconstruction performance across a range of visual data formats, suggesting the potential of SSMs for data-specific neural representations.

Recent attempt in dataset distillation has been made to compress large-scale training datasets into compact synthetic versions, significantly reducing memory usage and training costs. While parameterization-based approaches have shown promising results on image datasets, their application to 3D point clouds remains largely unexplored due to the irregular and unordered nature of 3D data. In this paper, we first introduce a parameterization-based dataset distillation framework for 3D point clouds that enables the use of more diverse synthetic samples than conventional methods under the same memory budget. We construct an initial synthetic dataset containing multiple anchor samples with a coarser resolution than the original sample. We also generate new samples by morphing the shapes of the anchor samples with learnable weights to improve the diversity of the synthetic dataset. Moreover, we devise a uniformity-aware matching loss to ensure the structural consistency when comparing the original and synthetic datasets. Extensive experiments conducted on five standard benchmarks—ModelNet10, ModelNet40, ShapeNet, ScanObjectNN, and OmniObject3D—demonstrate that the proposed method effectively optimizes both the synthetic samples and the weights for shape morphing, outperforming existing dataset distillation methods.

Object counting is a fundamental task in computer vision, with broad applicability in many real-world scenarios. Fully-supervised counting methods require costly point-level annotations per object. Few weakly-supervised methods leverage only image-level object counts as supervision and achieve fairly promising results. They are, however, often limited to counting a single category, \eg person. In this paper, we propose WS-COC, the first MLLM-driven weakly-supervised framework for class-agnostic object counting. Instead of directly fine-tuning MLLMs to predict object counts, which can be challenging due to the modality gap, we incorporate three simple yet effective strategies to bootstrap the counting paradigm in both training and testing: First, a divide-and-discern dialogue tuning strategy is proposed to guide the MLLM to determine whether the object count falls within a specific range and progressively break down the range through multi-round dialogue. Second, a compare-and-rank count optimization strategy is introduced to train the MLLM to optimize the relative ranking of multiple images according to their object counts. Third, a global-and-local counting enhancement strategy aggregates and fuses local and global count predictions to improve counting performance in dense scenes. Extensive experiments on FSC-147, CARPK, PUCPR+, and ShanghaiTech show that WS-COC matches or even surpasses many state-of-art fully-supervised methods while significantly reducing annotation costs.

Emotional Image Content Generation (EICG) aims to generate semantically clear and emotionally faithful images based on given emotion categories, with broad application prospects. While recent text-to-image diffusion models excel at generating concrete concepts, they struggle with the complexity of abstract emotions. There have also emerged methods specifically designed for EICG, but they excessively rely on word-level attribute labels for guidance, which suffer from semantic incoherence, ambiguity, and limited scalability. To address these challenges, we propose CoEmoGen, a novel pipeline notable for its semantic coherence and high scalability. Specifically, leveraging multimodal large language models (MLLMs), we construct high-quality captions focused on emotion-triggering content for context-rich semantic guidance. Furthermore, inspired by psychological insights, we design a Hierarchical Low-Rank Adaptation (HiLoRA) module to cohesively model both polarity-shared low-level features and emotion-specific high-level semantics. Extensive experiments demonstrate CoEmoGen’s superiority in emotional faithfulness and semantic coherence from quantitative, qualitative, and user study perspectives. To intuitively showcase scalability, we curate EmoArt, a large-scale dataset of emotionally evocative artistic images, providing endless inspiration for emotion-driven artistic creation. The dataset and code are available at https://github.com/yuankaishen2001/CoEmoGen.

With the rapid progress of large language models (LLMs), multimodal frameworks that unify understanding and generation have become promising, yet they face increasing complexity as the number of modalities and tasks grows. We observe that motion quantization introduces approximation errors that cap motion quality, and that unifying discrete text and continuous motion within a single-stream backbone amplifies cross-modal interference. Motivated by recent multi-branch Transformer designs that separate signals from different modalities, we propose MotionGPT3, a bimodal motion–language model for both understanding and generation. MotionGPT3 encodes raw motion into a continuous latent space using a variational autoencoder (VAE), thereby avoiding quantization-induced artifacts, while leveraging the semantic prior of pretrained language models. A dual-stream Transformer with shared attention preserves modality-specific routes while enabling controlled, bidirectional information flow, which reduces interference, stabilizing optimization, and empirically accelerates convergence without degrading fidelity. For multimodal joint training, a generate-then-align three-stage schedule further improves stability and limits cross-task interference. Experiments show that MotionGPT3 achieves 2× faster convergence in training loss and up to 4× faster convergence in validation, while maintaining state-of-the-art performance on standard motion understanding and motion generation benchmarks.

3D anime character generation has become increasingly important in digital entertainment, including animation production, virtual reality, gaming, and virtual influencers. Unlike realistic human modeling, anime-style characters require exaggerated proportions, stylized surface details, and artistically consistent garments, posing unique challenges for automated 3D generation. Previous approaches for 3D anime character generation often suffer from low mesh quality and blurry textures, and they typically do not provide corresponding skeletons, limiting their usability in animation. In this work, we present a novel framework for high-quality 3D anime character generation that overcomes these limitations by combining the expressive power of the Skinned Multi-Person Linear (SMPL) model with precise garment generation. Our approach extends the Anime-SMPL model to better capture the distinct features of anime characters, enabling unified skeleton generation and blendshape-based facial expression control. This results in fully animation-ready 3D characters with expressive faces, bodies, and garments. To complement the body model, we introduce a body-aligned component-wise garments generation pipeline (including hairstyles, upper garments, lower garments, and accessories), which models garments as structured components aligned with body geometry. Furthermore, our method produces high-quality skin and facial textures, as well as detailed garment textures, enhancing the visual fidelity of the generated characters. Experimental results demonstrate that our framework significantly outperforms baseline methods in terms of mesh quality, texture clarity, and garment-body alignment, making it suitable for a wide range of applications in anime content creation and interactive media.

Dense semantic segmentation in dynamic environments is fundamentally limited by the low-frame-rate (LFR) nature of standard cameras, which creates critical perceptual gaps between frames. To solve this, we introduce Anytime Interframe Semantic Segmentation: a new task for predicting segmentation at any arbitrary time using only a single past RGB frame and a stream of asynchronous event data. This task presents a core challenge: how to robustly propagate dense semantic features using a motion field derived from sparse and often noisy event data, all while mitigating feature degradation in highly dynamic scenes. We propose LiFR-Seg, a novel framework that directly addresses these challenges by propagating deep semantic features through time. The core of our method is an uncertainty-aware warping process, guided by an event-driven motion field and its learned, explicit confidence. A temporal memory attention module further ensures coherence in dynamic scenarios. We validate our method on the DSEC dataset and a new high-frequency synthetic benchmark (SHF-DSEC) we contribute. Remarkably, our LFR system achieves performance (73.82\% mIoU on DSEC) that is statistically indistinguishable from an HFR upper-bound (within 0.09\%) that has full access to the target frame. % We further demonstrate superior robustness in highly dynamic (M3ED-Drone & Quadruped) and low-light (DSEC-Night) scenarios, where our method can even surpass the HFR baseline. We further demonstrate superior robustness across extreme scenarios: in highly dynamic (M3ED) tests, our method closely matches the HFR baseline's performance, while in the low-light (DSEC-Night) evaluation, it even surpasses it. This work presents a new, efficient paradigm for achieving robust, high-frame-rate perception with low-frame-rate hardware.

Visual prompted object detection enables interactive and flexible definition of target categories, thereby facilitating open-vocabulary detection. Since visual prompts are derived directly from image features, they often outperform text prompts in recognizing rare categories. Nevertheless, research on visual prompted detection has been largely overlooked, and it is typically treated as a byproduct of training text prompted detectors, which hinders its development. To fully unlock the potential of visual-prompted detection, we investigate the reasons why its performance is suboptimal and reveal that the underlying issue lies in the absence of global discriminability in visual prompts. Motivated by these observations, we propose DETR-ViP, a robust object detection framework that yields class-distinguishable visual prompts. On top of basic image-text contrastive learning, DETR-ViP incorporates global prompt integration and visual-textual prompt relation distillation to learn more discriminative prompt representations. In addition, DETR-ViP employs a selective fusion strategy that ensures stable and robust detection. Extensive experiments on COCO, LVIS, ODinW, and Roboflow100 demonstrate that DETR-ViP achieves substantially higher performance in visual prompt detection compared to other state-of-the-art counterparts. A series of ablation studies and analyses further validate the effectiveness of the proposed improvements and shed light on the underlying reasons for the enhanced detection capability of visual prompts.

The rapid deployment of large language models (LLMs) in consumer applications has led to frequent exchanges of personal information. To obtain useful responses, users often share more than necessary, increasing privacy risks via memorization, context-based personalization, or security breaches. We present a framework to formally define and operationalize data minimization: for a given user prompt and response model, quantifying the least privacy-revealing disclosure that maintains utility, and propose a priority-queue tree search to locate this optimal point within a privacy-ordered transformation space. We evaluated the framework on four datasets spanning open-ended conversations (ShareGPT, WildChat) and knowledge-intensive tasks with single-ground-truth answers (CaseHOLD, MedQA), quantifying achievable data minimization with nine LLMs as the response model. Our results demonstrate that larger frontier LLMs can tolerate stronger data minimization while maintaining task quality than smaller open-source models (85.7% redaction for GPT-5 vs. 19.3% for Qwen2.5-0.5B). By comparing with our search-derived benchmarks, we find that LLMs struggle to predict optimal data minimization directly, showing a bias toward abstraction that leads to oversharing. This suggests not just a privacy gap, but a capability gap: models may lack awareness of what information they actually need to solve a task.

Assessing the privacy of large language models (LLMs) presents significant challenges. In particular, most existing methods for auditing differential privacy require the insertion of specially crafted canary data during training, making them impractical for auditing already-trained models without costly retraining. Additionally, dataset inference, which audits whether a suspect dataset was used to train a model, is infeasible without access to a private non-member held-out dataset. Yet, such held-out datasets are often unavailable or difficult to construct for real-world cases since they have to be from the same distribution (IID) as the suspect data. These limitations severely hinder the ability to conduct scalable, post-hoc audits. To enable such audits, this work introduces natural identifiers (NIDs) as a novel solution to the above-mentioned challenges. NIDs are structured random strings, such as cryptographic hashes and shortened URLs, naturally occurring in common LLM training datasets. Their format enables the generation of unlimited additional random strings from the same distribution, which can act as alternative canaries for audits and as same-distribution held-out data for dataset inference. Our evaluation highlights that indeed, using NIDs, we can facilitate post-hoc differential privacy auditing without any retraining and enable dataset inference for any suspect dataset containing NIDs without the need for a private non-member held-out dataset.

Activation steering has emerged as a cost-effective paradigm for modifying large language model (LLM) behaviors. Existing methods typically intervene at the block level, steering the bundled activations of selected attention heads, feedforward networks, or residual streams. However, we reveal that block-level activations are inherently heterogeneous, entangling beneficial, irrelevant, and harmful features, thereby rendering block-level steering coarse, inefficient, and intrusive. To investigate the root cause, we decompose block activations into fine-grained atomic unit (AU)–level activations, where each AU-level activation corresponds to a single dimension of the block activation, and each AU denotes a slice of the block weight matrix. Steering an AU-level activation is thus equivalent to steering its associated AU. Our theoretical and empirical analysis show that heterogeneity arises because different AUs or dimensions control distinct token distributions in LLM outputs. Hence, block-level steering inevitably moves helpful and harmful token directions together, which reduces efficiency. Restricting intervention to beneficial AUs yields more precise and effective steering. Building on this insight, we propose AUSteer, a simple and efficient method that operates at a finer granularity of the AU level. AUSteer first identifies discriminative AUs globally by computing activation momenta on contrastive samples. It then assigns adaptive steering strengths tailored to diverse inputs and selected AU activations. Comprehensive experiments on multiple LLMs and tasks show that AUSteer consistently surpasses advanced baselines while steering considerably fewer activations, demonstrating that steering less achieves more.

Training machine learning models requires the storage of large datasets, which often contain sensitive or private data. Storing data is associated with a number of potential risks which increase over time, such as database breaches and malicious adversaries. Machine unlearning is the study of methods to efficiently remove the influence of training data subsets from previously-trained models. Existing unlearning methods typically require direct access to the "forget set"---the data to be forgotten-and organisations must retain this data for unlearning rather than deleting it immediately upon request, increasing risks associated with the forget set. We introduce partially-blind unlearning---utilizing auxiliary information to unlearn without explicit access to the forget set. We also propose a practical framework Reload, a partially-blind method based on gradient optimization and structured weight sparsification to operationalize partially-blind unlearning. We show that Reload efficiently unlearns, approximating models retrained from scratch, and outperforms several forget set-dependent approaches. On language models, Reload unlearns entities using <0.025\% of the retain set and <7\% of model weights in <8 minutes on Llama2-7B. In the corrective case, Reload achieves unlearning even when only 10\% of corrupted data is identified.

Secure multiparty computation allows the client to secretly inference their sensitive inputs without acquiring the proprietary machine learning model weights. As the decoder-only transformer-based large language model becomes the popular paradigm, the desire of applying MPC in large language models is increasing. However, such inference usually leads to great amount of latency, which is due to nonlinear operations in the Transformer architecture. Recent works either focus on improving cryptographic primitives or re-architecting and re-training to make LLM MPC-friendly. We, on the other hand, observe that properly addressing outlier phenomena, which are unique yet universal properties existing across different LLMs, can effectively reduce the input domain and thereby design faster protocols for non-linear operations. Hence, we propose Secure Outlier-Aware Large Language Model Inference framework (SOAL), which accelerates the RMSNorm operation by nearly 2 $\times$, SiLU by $2\times$, and Softmax by more than 5$\times$. SOAL maintains the same performance of the original model without any fine-tuning requirement.

Large-scale deep learning models are known to *memorize* parts of the training set. In machine learning theory, memorization is often framed as interpolation or label fitting, and classical results show that this can be achieved when the number of parameters $p$ in the model is larger than the number of training samples $n$. In this work, we consider memorization from the perspective of *data reconstruction*, demonstrating that this can be achieved when $p$ is larger than $dn$, where $d$ is the dimensionality of the data. More specifically, we show that, in the random features model, when $p \gg dn$, the subspace spanned by the training samples in feature space gives sufficient information to identify the individual samples in input space. Our analysis suggests an optimization method to reconstruct the dataset from the model parameters, and we demonstrate that this method performs well on various architectures (random features, two-layer fully-connected and deep residual networks). Our results reveal a *law of data reconstruction*, according to which the entire training dataset can be recovered as $p$ exceeds the threshold $dn$.

The pre-training of large language models (LLMs) relies on massive text datasets sourced from diverse and difficult-to-curate origins. Although membership inference attacks and hidden canaries have been explored to trace data usage, such methods rely on *regurgitation* of training data, which LM providers try to limit. In this work, we demonstrate that *indirect data poisoning* (where the targeted behavior is absent from training data) is not only feasible against LLMs but also allows to effectively protect a dataset and trace its use. Using gradient-based optimization prompt-tuning, we craft poisons to make a model learn arbitrary *secret sequences*: secret responses to secret prompts that are **absent from the training corpus**.\ We validate our approach on language models pre-trained from scratch and show that less than 0.005\% of poisoned tokens are sufficient to covertly make a LM learn a *secret* and detect it with extremely high confidence ( $p < 10^{-55}$ ) with a theoretically certifiable scheme. Crucially, this occurs without performance degradation (on LM benchmarks) and despite secrets **never appearing in the training set**.

Fully polynomial neural networks—models whose computations comprise only additions and multiplications—are attractive for privacy-preserving inference under homomorphic encryption (HE). Yet most prior systems obtain such models by post-hoc replacement of nonlinearities with high-degree or cascaded polynomials, which inflates HE cost and makes training numerically fragile and hard to scale. We introduce ULD-Net, a training methodology that enables ultra-low-degree (multiplicative depth $\leq 3$ for each operator) fully polynomial networks to be trained from scratch at ImageNet and transformer scale while maintaining high accuracy. The key is a polynomial-only normalization, PolyNorm, coupled with a principled choice of normalization axis that keeps activations in a well-conditioned range across deep stacks of polynomial layers. Together with a special set of polynomial-aware operator replacements, such as polynomial activation functions and linear attention, ULD-Net delivers stable optimization without resorting to high-degree approximations. Experimental results demonstrate that ULD-Net enables stable training of low-degree fully polynomial networks on large-scale model architectures and datasets. Applying ULD-Net to ViT-Small and ViT-Base achieves 76.70\% and 75.20\% top-1 accuracy on ImageNet, respectively, which are comparable to the original models and represent the first fully polynomial models successfully scaled to the ViT/ImageNet level. Additionally, ULD-Net outperforms several state-of-the-art open-source fully and partially polynomial approaches across diverse model architectures and datasets in both accuracy and HE inference latency.

Differential privacy (DP) is widely employed in machine learning to protect confidential or sensitive training data from being revealed. As data owners gain greater control over their data due to personal data ownership, they are more likely to set their own privacy requirements, necessitating individualized DP (IDP) to fulfil such requests. In particular, owners of data from more sensitive subsets, such as positive cases of stigmatized diseases, likely set stronger privacy requirements, as leakage of such data could incur more serious societal impact. However, existing IDP algorithms induce a critical utility imbalance problem: Data from owners with stronger privacy requirements may be severely underrepresented in the trained model, resulting in poorer performance on similar data from subsequent users during deployment. In this paper, we analyze this problem and propose the INO-SGD algorithm, which strategically down-weights data within each batch to improve performance on the more private data across all iterations. Notably, our algorithm is specially designed to satisfy IDP, while existing techniques addressing utility imbalance neither satisfy IDP nor can be easily adapted to do so. Lastly, we demonstrate the empirical feasibility of our approach.

Matrix factorization mechanisms for differentially private training have emerged as a promising approach to improve model utility under privacy constraints. In practical settings, models are typically trained over multiple epochs, requiring matrix factorizations that account for repeated participation. Existing theoretical upper and lower bounds on multi-epoch factorization error leave a significant gap. In this work, we introduce a new explicit factorization method, Banded Inverse Square Root (BISR), which imposes a banded structure on the inverse correlation matrix. This factorization enables us to derive an explicit and tight characterization of the multi-epoch error. We further prove that BISR achieves asymptotically optimal error by matching the upper and lower bounds. Empirically, BISR performs on par with the state of the art factorization methods, while being simpler to implement, computationally efficient, and easier to analyze.

The powerful cooperation of federated learning (FL) and differential privacy (DP) provides a promising paradigm for the large-scale private clients. However, existing analyses in FL-DP mostly rely on the composition theorem and cannot tightly quantify the privacy leakage challenges, which is tight for a few communication rounds but yields an arbitrarily loose and divergent bound eventually. This also implies a counterintuitive judgment, suggesting that FL-DP may not provide adequate privacy support during long-term training under constant-level noisy perturbations, yielding discrepancy between the theoretical and experimental results. To further investigate the convergent privacy and reliability of the FL-DP framework, in this paper, we comprehensively evaluate the worst privacy of two classical methods under the non-convex and smooth objectives based on the $f$-DP analysis. With the aid of the shifted interpolation technique, we successfully prove that privacy in Noisy-FedAvg has a tight convergent bound. Moreover, with the regularization of the proxy term, privacy in Noisy-FedProx has a stable constant lower bound. Our analysis further demonstrates a solid theoretical foundation for the reliability of privacy in FL-DP. Meanwhile, our conclusions can also be losslessly converted to other classical DP analytical frameworks, e.g. $(\epsilon,\delta)$-DP and R$\'{e}$nyi-DP (RDP), to provide more fine-grained understandings for the FL-DP frameworks.

As diffusion model-based services expand across various domains, safeguarding client data privacy has become increasingly critical. While fully homomorphic encryption and secure multi-party computation enable privacy-preserving inference, their high computational overhead poses challenges for large-scale diffusion applications. Recent work alleviates computational costs by substituting non-linear operations with low-degree polynomial approximations. While such relaxations reduce latency, they incur significant degradation in generative fidelity, and inference remains considerably slower than plaintext execution. To further accelerate secure inference while preserving performance, we explore more relaxed approximations and propose a score-correction framework that rectifies the conditional score shift induced by the relaxed approximation, rather than decreasing the approximation error itself. The key insight is that unconditional generation can be executed without approximation and thus provides a high-fidelity score signal. Leveraging this unconditional score as corrective guidance enables more relaxed approximations while preserving semantic and perceptual quality. In experiments, we demonstrate that our method significantly alleviates the performance degradation caused by relaxed approximations across various benchmarks.

Federated Learning (FL) was initially proposed as a privacy-preserving machine learning paradigm. However, FL has been shown to be susceptible to a series of privacy attacks. Recently, there has been concern about the Source Inference Attack (SIA), where an honest-but-curious central server attempts to identify exactly which client owns a given data point which was used in the training phase. Alarmingly, standard gradient obfuscation techniques with Differential Privacy have been shown to be ineffective against SIAs, at least without severely diminishing the accuracy. In this work, we propose a defense against SIAs within the widely studied shuffle model of FL, where an honest shuffler acts as an intermediary between the clients and the server. First, we demonstrate that standard naive shuffling alone is insufficient to prevent SIAs. To effectively defend against SIAs, shuffling needs to be applied at a more granular level; we propose a novel combination of parameter-level shuffling with the residue number system (RNS). Our approach provides robust protection against SIAs without affecting the accuracy of the joint model and can be seamlessly integrated into other privacy protection mechanisms. We conduct experiments on a series of models and datasets, confirming that standard shuffling approaches fail to prevent SIAs and that, in contrast, our proposed method reduces the attack’s accuracy to the level of random guessing.

Machine unlearning has emerged as an inevitable AI mechanism to support GDPR requirements such as revoking user consent through the "right to be forgotten". However, existing approaches often leave residual traces that make them vulnerable to data reconstruction attacks. In this work, we propose ReTrace, the first reconstruction attack framework that uniquely formulates unlearned data recovery on large-scale deep architectures as a reinforcement learning (RL) problem. By treating residual unlearning traces as reward signals, ReTrace guides a generator to actively explore the input space and converge toward the forgotten data distribution. This RL-guided approach enables both instance-level recovery of individual samples and distribution-level reconstruction of unlearned classes. We provide a theoretical foundation showing that the RL objective converges to an exponential-tilted distribution that amplifies forgotten regions. Empirically, ReTrace achieves up to 73.1\% instance-level recovery and reduces FID and KL scores beyond two state-of-the-art baselines. Strikingly, on the challenging task of text unlearning, it improves BLEU scores by nearly 100\% over black-box baselines while preserving distributional fidelity, demonstrating that RL can recover even high-dimensional and structured modalities. Furthermore, ReTrace demonstrates effectiveness across both convolutional (ResNet) and transformer-based models, with Distil-BERT as the largest architecture attacked to date. These results show that current unlearning methods remain vulnerable, highlighting the need for robust and provably private mechanisms.

Backdoor attacks pose a significant threat to machine learning models, allowing adversaries to implant hidden triggers that alter model behavior when activated. Although gradient ascent (GA)-based unlearning has been proposed as an efficient backdoor removal approach, we identify a critical yet overlooked issue: vanilla GA does not eliminate the trigger but shifts its impact to different classes, a phenomenon we call trigger shifting. To address this, we propose Robust Gradient Ascent (RGA), which introduces a dynamic penalty mechanism to regulate GA's strength and prevent excessive unlearning. Our experiments show that RGA effectively removes backdoors while preserving model utility, offering a more reliable defense against backdoor attacks.

Unlearnable Examples (UEs) serve as a data protection strategy that generates imperceptible perturbations to mislead models into learning spurious correlations instead of underlying semantics. In this paper, we uncover a fundamental vulnerability of UEs that emerges when learning starts from a pretrained model. Crucially, our empirical analysis shows that even when data are protected by carefully crafted perturbations, pretraining priors still furnish rich semantic representations that allow the model to circumvent the shortcuts introduced by UEs and capture genuine features, thereby nullifying unlearnability. To address this, we propose $\textbf{BAIT}$ ($\textbf{B}$inding $\textbf{A}$rtificial perturbations to $\textbf{I}$ncorrect $\textbf{T}$argets), a novel bi‑level optimization formulation. Specifically, the inner level aims at associating the perturbed samples with real labels to simulate standard data-label alignment, while the outer level actively disrupts this alignment by enforcing a mislabel-perturbation binding that maps samples to designated incorrect targets. This mechanism effectively overrides the semantic guidance of priors, forcing the model to rely on the injected perturbations and consequently preventing the acquisition of true semantics. Extensive experiments on standard benchmarks and multiple pretrained backbones demonstrate that BAIT effectively mitigates the influence of pretraining priors and maintains data unlearnability. Code is available at https://github.com/zhli-cs/BAIT.

Adversarial-example-based fingerprinting approaches, which leverage the decision boundary characteristics of deep neural networks (DNNs) to craft fingerprints, has proven effective for protecting model ownership. However, a fundamental challenge remains unresolved: how far a fingerprint should be placed from the decision boundary to simultaneously satisfy two essential properties—robustness and uniqueness—required for effective and reliable ownership protection. Despite the importance of the fingerprint-to-boundary distance, existing works offer no theoretical solution and instead rely on empirical heuristics to determine it, which may lead to violations of either robustness or uniqueness properties. We propose AnaFP, an analytical fingerprinting scheme that constructs fingerprints under theoretical guidance. Specifically, we formulate the fingerprint generation task as the problem of controlling the fingerprint-to-boundary distance through a tunable stretch factor. To ensure both robustness and uniqueness, we mathematically formalize these properties that determine the lower and upper bounds of the stretch factor. These bounds jointly define an admissible interval within which the stretch factor must lie, thereby establishing a theoretical connection between the two constraints and the fingerprint-to-boundary distance. To enable practical fingerprint generation, we approximate the original (infinite) sets of pirated and independently trained models using two finite surrogate model pools and employ a quantile-based relaxation strategy to relax the derived bounds. Particularly, due to the circular dependency between the lower bound and the stretch factor, we apply a grid search strategy over the admissible interval to determine the most feasible stretch factor. Extensive experimental results demonstrate that AnaFP consistently outperforms prior methods, achieving effective and reliable ownership verification across diverse model architectures and model modification attacks.

Deep neural networks are known to often memorize underrepresented, hard examples, with implications for generalization and privacy. Feldman & Zhang (2020) defined a rigorous notion of memorization. However it is prohibitively expensive to compute at scale because it requires training models both with and without the data point of interest in order to calculate the memorization score. We observe that samples that are less memorized tend to be learned earlier in training, whereas highly memorized samples are learned later. Motivated by this observation, we introduce Cumulative Sample Gradient (CSG), a computationally efficient proxy for memorization. CSG is the gradient of the loss with respect to input samples, accumulated over the course of training. The advantage of using input gradients is that per-sample gradients can be obtained with negligible overhead during training. The accumulation over training also reduces per-epoch variance and enables a formal link to memorization. Theoretically, we show that CSG is bounded by memorization and by learning time. Tracking these gradients during training reveals a characteristic rise–peak–decline trajectory whose timing is mirrored by the model’s weight norm. This yields an early-stopping criterion that does not require a validation set: stop at the peak of the weight norm. This early stopping also enables our memorization proxy, CSG, to be up to five orders of magnitude more efficient than the memorization score from Feldman & Zhang (2020). It is also approximately 140 $\times$ and 10$\times$ faster than the prior state-of-the-art memorization proxies, input curvature and cumulative sample loss, while still aligning closely with the memorization score, exhibiting high correlation. Further, we develop Sample Gradient Assisted Loss (SGAL), a proxy that further improves alignment with memorization and is highly efficient to compute. Finally, we show that CSG attains state-of-the-art performance on practical dataset diagnostics, such as mislabeled-sample detection and enables bias discovery, providing a theoretically grounded toolbox for studying memorization in deep networks.

In recent years, diffusion models (DMs) have achieved remarkable success in recommender systems (RSs), owing to their strong capacity to model the complex distributions of item content and user behaviors. Despite their effectiveness, existing methods pose the danger of generating uncredible content recommendations (e.g., fake news, misinformation) that may significantly harm social well-being, as they primarily emphasize recommendation accuracy while neglecting the credibility of the recommended content. To address this issue, in this paper, we propose Disco, a novel method to steer diffusion models towards credible content recommendation. Specifically, we design a novel disentangled diffusion model to mitigate the harmful influence of uncredible content on the generation process while preserving high recommendation accuracy. This is achieved by reformulating the diffusion objective to encourage generation conditioned on preference-related signals while discouraging generation conditioned on uncredible content-related signals. In addition, to further improve the recommendation credibility, we design a progressively enhanced credible subspace projection that suppresses uncredible content by projecting diffusion targets into the null space of uncredible content. Extensive experiments on real-world datasets demonstrate the effectiveness of Disco in terms of both accurate and credible content recommendations.

As Retrieval-Augmented Generation (RAG) evolves into service-oriented platforms (Rag-as-a-Service) with shared knowledge bases, protecting the copyright of contributed data becomes essential. Existing watermarking methods in RAG focus solely on textual knowledge, leaving image knowledge unprotected. In this work, we propose \textit{AQUA}, the first watermark framework for image knowledge protection in Multimodal RAG systems. \textit{AQUA} embeds semantic signals into synthetic images using two complementary methods: acronym-based triggers and spatial relationship cues. These techniques ensure watermark signals survive indirect watermark propagation from image retriever to textual generator, being efficient, effective and imperceptible. Experiments across diverse models and datasets show that \textit{AQUA} enables robust, stealthy, and reliable copyright tracing, filling a key gap in multimodal RAG protection.

Machine unlearning (MU) for large language models (LLMs), commonly referred to as LLM unlearning, seeks to remove specific undesirable data or knowledge from a trained model, while maintaining its performance on standard tasks. While unlearning plays a vital role in protecting data privacy, enforcing copyright, and mitigating sociotechnical harms in LLMs, we identify a new vulnerability post-unlearning: unlearning trace detection. We discover that unlearning leaves behind persistent "fingerprints" in LLMs, detectable traces in both model behavior and internal representations. These traces can be identified from output responses, even when prompted with forget-irrelevant inputs. Specifically, even a simple supervised classifier can determine whether a model has undergone unlearning, using only its prediction logits or even its textual outputs. Further analysis shows that these traces are embedded in intermediate activations and propagate nonlinearly to the final layer, forming low-dimensional, learnable manifolds in activation space. Through extensive experiments, we demonstrate that unlearning traces can be detected with over 90% accuracy even under forget-irrelevant inputs, and that larger LLMs exhibit stronger detectability. These findings reveal that unlearning leaves measurable signatures, introducing a new risk of reverse-engineering forgotten information when a model is identified as unlearned, given an input query.

Large language models (LLMs) can exhibit concept-conditioned semantic divergence: common high-level cues (e.g., ideologies, public figures) elicit unusually uniform, stance-like responses that evade token-trigger audits. This behavior falls in a blind spot of current safety evaluations, yet carries major societal stakes, as such concept cues can steer content exposure at scale. We formalize this phenomenon and present RAVEN (Response Anomaly Vigilance), a black-box audit that flags cases where a model is simultaneously highly certain and atypical among peers by coupling semantic entropy over paraphrastic samples with cross-model disagreement. In a controlled LoRA fine-tuning study, we implant a concept-conditioned stance using a small biased corpus, demonstrating feasibility without rare token triggers. Auditing five LLM families across twelve sensitive topics (360 prompts per model) and clustering via bidirectional entailment, RAVEN surfaces recurrent, model-specific divergences in 9/12 topics. Concept-level audits complement token-level defenses and provide a practical early-warning signal for release evaluation and post-deployment monitoring against propaganda-like influence.

Large language models remain vulnerable to jailbreak attacks that bypass safety guardrails to elicit harmful outputs. Defending against novel jailbreaks represents a critical challenge in AI safety. Adversarial training---designed to make models robust against worst-case perturbations---has been the dominant paradigm for adversarial robustness. However, due to optimization challenges and difficulties in defining realistic threat models, adversarial training methods often fail on newly developed jailbreaks in practice. This paper proposes a new paradigm for improving robustness against unseen jailbreaks, centered on the Adversarial Déjà Vu hypothesis: novel jailbreaks are not fundamentally new, but largely recombinations of adversarial skills from previous attacks. We study this hypothesis through a large-scale analysis of 32 attack papers published over two years. Using an automated pipeline, we extract and compress adversarial skills into a sparse dictionary of primitives, with LLMs generating human-readable descriptions. Our analysis reveals that unseen attacks can be effectively explained as sparse compositions of earlier skills, with explanatory power increasing monotonically as skill coverage grows. Guided by this insight, we introduce Adversarial Skill Compositional Training (ASCoT), which trains on diverse compositions of skill primitives rather than isolated attack instances. ASCoT substantially improves robustness to unseen attacks, including multi-turn jailbreaks, while maintaining low over-refusal rates. We also demonstrate that expanding adversarial skill coverage, not just data scale, is key to defending against novel attacks.

Cybersecurity operations demand assistant LLMs that support diverse workflows without exposing sensitive data. Existing solutions either rely on proprietary APIs with privacy risks or on open models lacking domain adaptation. To bridge this gap, we curate 11.8B tokens of cybersecurity-focused continual pretraining data via large-scale web filtering and manual collection of high-quality resources, spanning 28.6K documents across frameworks, offensive techniques, and security tools. Building on this, we design an agentic augmentation pipeline that simulates expert workflows to generate 266K multi-turn cybersecurity samples for supervised fine-tuning. Combined with general open-source LLM data, these resources enable the training of RedSage, an open-source, locally deployable cybersecurity assistant with domain-aware pretraining and post-training. To rigorously evaluate the models, we introduce RedSage-Bench, a benchmark with 30K multiple-choice and 240 open-ended Q&amp;A items covering cybersecurity knowledge, skills, and tool expertise. RedSage is further evaluated on established cybersecurity benchmarks (e.g., CTI-Bench, CyberMetric, SECURE) and general LLM benchmarks to assess broader generalization. At the 8B scale, RedSage achieves consistently better results, surpassing the baseline models by up to +5.59 points on cybersecurity benchmarks and +5.05 points on Open LLM Leaderboard tasks. These findings demonstrate that domain-aware agentic augmentation and pre/post-training can not only enhance cybersecurity-specific expertise but also help to improve general reasoning and instruction-following. Project page: https://risys-lab.github.io/RedSage/

Safety preferences are inherently subjective, yet current LLM safety alignment methods often impose universal standards that fail to account for individual sensitivities. In this work, we propose an efficient, training-free method for personalized safety control via inference-time activation intervention. Our approach steers internal representations to suppress user-specific undesired content while preserving model utility. We systematically evaluate three strategies for estimating intervention directions: Instance-Level Contrast Shift (ILCS), Unpaired Mean Shift (UMS), and our primary method, Paired Contrast Mean Shift (PCMS). We provide theoretical insights into each approach and highlight the advantages of PCMS. Empirical results across diverse open-weight models demonstrate that our method effectively reduces undesired content in line with individual preferences, with minimal impact on helpfulness—enabling more adaptive and user-aligned LLM behavior.

Unlearning has emerged as a key technique to mitigate harmful content generation in diffusion models. However, existing methods often remove not only the target concept, but also benign co-occurring concepts. Unlearning nudity can unintentionally suppress the concept of person, preventing a model from generating images with person. We define these undesirably suppressed co-occurring concepts that must be preserved $\textbf{CARE}$ ($\textbf{C}$o-occurring $\textbf{A}$ssociated $\textbf{RE}$tained concepts). Then, we introduce the $\textbf{CARE score}$, a general metric that directly quantifies their preservation across unlearning tasks. With this foundation, we propose $\textbf{ReCARE}$ ($\textbf{R}$obust $\textbf{e}$rasure for $\textbf{CARE}$), a framework that explicitly safeguards CARE while erasing only the target concept. ReCARE automatically constructs the CARE-set, a curated vocabulary of benign co-occurring tokens extracted from target images, and leverages this vocabulary during training for stable unlearning. Extensive experiments across various target concepts ($\textit{Nudity}$, $\textit{Van Gogh}$ style, and $\textit{Tench}$ object) demonstrate that ReCARE achieves overall state-of-the-art performance in balancing robust concept erasure, overall utility, and CARE preservation.

Backdoor attacks pose severe security threats to large language models (LLMs), where a model behaves normally under benign inputs but produces malicious outputs when a hidden trigger appears. Existing backdoor removal methods typically assume prior knowledge of triggers, access to a clean reference model, or rely on aggressive finetuning configurations, and are often limited to classification tasks. However, such assumptions fall apart in real-world generative LLM settings. In this work, we propose a new framework for purifying generative LLM without any prior trigger knowledge or clean references. Through systematic sanity checks, we find that backdoor associations are redundantly encoded across MLP layers, while attention modules primarily amplify trigger signals without establishing the behavior. Leveraging this insight, we shift the focus from isolating specific backdoor triggers to cutting off the trigger–behavior associations, and design an immunization-inspired elimination approach: by constructing multiple synthetic backdoored variants of the given suspicious model, each trained with different malicious trigger–behavior pairs, and contrasting them with their clean counterparts. The recurring modifications across variants reveal a shared "backdoor signature"—analogous to antigens in a virus. Guided by this signature, we neutralize highly suspicious components in LLM and apply lightweight finetuning to restore its fluency, producing purified models that withstand diverse backdoor attacks and threat models while preserving generative capability.

While machine-generated texts (MGTs) offer great convenience, they also pose risks such as disinformation and phishing, highlighting the need for reliable detection. Metric-based methods, which extract statistically distinguishable features of MGTs, are often more practical than complex model-based methods that are prone to overfitting. Given their diverse designs, we first place representative metric-based methods within a unified framework, enabling a clear assessment of their advantages and limitations. Our analysis identifies a core challenge across these methods: the token-level detection score is easily biased by the inherent randomness of the MGTs generation process. To address this, we theoretically and empirically reveal two relationships of context detection scores that may aid calibration: Neighbor Similarity and Initial Instability. We then propose a Markov-informed score calibration strategy that models these relationships using Markov random fields, and implements it as a lightweight component via a mean-field approximation, allowing our method to be seamlessly integrated into existing detectors. Extensive experiments in various real-world scenarios, such as cross-LLM and paraphrasing attacks, demonstrate significant gains over baselines with negligible computational overhead. The code is available at \url{ https://github.com/tmlr-group/MRF_Calibration}.

Large Language Models (LLMs) are trained with safety alignment to prevent generating malicious content. Although some attacks have highlighted vulnerabilities in these safety-aligned LLMs, they typically have limitations, such as necessitating access to the model weights or the generation process. Since proprietary models through API-calling do not grant users such permissions, these attacks find it challenging to compromise them. In this paper, we propose Jailbreaking Using LLM Introspection (JULI), which jailbreaks LLMs by manipulating the token log probabilities, using a tiny plug-in block, BiasNet. JULI relies solely on the knowledge of the target LLM's predicted token log probabilities. It can effectively jailbreak API-calling LLMs under a black-box setting and knowing only top-$5$ token log probabilities. Our approach demonstrates superior effectiveness, outperforming existing state-of-the-art (SOTA) approaches across multiple metrics.

The volume of freely scraped data on the Internet has driven the tremendous success of deep learning. Along with this comes the rising concern about data privacy and security. Numerous methods for generating unlearnable examples have been proposed to prevent data from being illicitly learned by unauthorized deep models by impeding generalization. However, the existing approaches primarily rely on empirical heuristics, making it challenging to enhance unlearnable examples with solid explanations. In this paper, we analyze and improve unlearnable examples from a novel perspective: mutual information reduction. We demonstrate that effective unlearnable examples always decrease mutual information between clean features and poisoned features, and when the network gets deeper, the unlearnability goes better together with lower mutual information. Further, we prove from a covariance reduction perspective that minimizing the conditional covariance of intra-class poisoned features reduces the mutual information between distributions. Based on the theoretical results, we propose a novel unlearnable method called Mutual Information Unlearnable Examples (MI-UE) that reduces covariance by maximizing the cosine similarity among intra-class features, thus impeding the generalization effectively. Extensive experiments demonstrate that our approach significantly outperforms previous state-of-the-art methods, even under defense mechanisms.

The objectives that Large Language Models (LLMs) implicitly optimize remain dangerously opaque, making trustworthy alignment and auditing a grand challenge. While Inverse Reinforcement Learning (IRL) can infer reward functions from behaviour, existing approaches either produce a single, overconfident reward estimate or fail to address the fundamental ambiguity of the task (non-identifiability). This paper introduces a principled auditing framework that re-frames reward inference from a simple estimation task to a comprehensive process for verification. Our framework leverages Bayesian IRL to not only recover a distribution over objectives but to enable three critical audit capabilities: (i) Quantifying and systematically reducing non-identifiability by demonstrating posterior contraction over sequential rounds of evidence; (ii) Providing actionable, uncertainty-aware diagnostics that expose spurious shortcuts and identify out-of-distribution prompts where the inferred objective cannot be trusted; and (iii) Validating policy-level utility by showing that the refined, low-uncertainty reward can be used directly in RLHF to achieve training dynamics and toxicity reductions comparable to the ground-truth alignment process. Empirically, our framework successfully audits a detoxified LLM and generalizes beyond detoxification to a helpfulness preference setting, yielding a well-calibrated and interpretable objective that strengthens alignment guarantees. Overall, this work provides a practical toolkit for auditors, safety teams, and regulators to verify what LLMs are truly trying to achieve, moving us toward more trustworthy and accountable AI.

We characterize the effectiveness of Sharpness-aware minimization (SAM) under machine unlearning scheme, where unlearning forget signals interferes with learning retain signals. While previous work prove that SAM improves generalization with noise memorization prevention, we show that SAM abandons such denoising property when fitting the forget set, leading to altered generalization depending on signal strength. We further characterize the signal surplus of SAM in the order of signal strength, which enables learning from less retain signals to maintain model performance and putting more weight on unlearning the forget set. Empirical studies show that SAM outperforms SGD with relaxed requirement for retain signals and can enhance various unlearning methods either as pretrain or unlearn algorithm. Motivated by our refined characterization of SAM unlearning and observing that overfitting can benefit more stringent sample-specific unlearning, we propose Sharp MinMax, which splits the model into two to learn retain signals with SAM and unlearn forget signals with sharpness maximization, achieving best performance. Extensive experiments show that SAM enhances unlearning across varying difficulties measured by memorization, yielding decreased feature entanglement between retain and forget sets, stronger resistance to membership inference attacks, and a flatter loss landscape. Our observations generalize to more noised data, different optimizers, and different architectures.

Large language models trained on web-scale data can memorize private or sensitive knowledge, raising significant privacy risks. Although some unlearning methods mitigate these risks, they remain vulnerable to "relearning" during subsequent training, allowing a substantial portion of forgotten knowledge to resurface. In this paper, we show that widely used unlearning methods cause shallow alignment: instead of faithfully erasing target knowledge, they generate spurious unlearning neurons that amplify negative influence to hide it. To overcome this limitation, we introduce Ssiuu, a new class of unlearning methods that employs attribution-guided regularization to prevent spurious negative influence and faithfully remove target knowledge. Experimental results confirm that our method reliably erases target knowledge and outperforms strong baselines across two practical retraining scenarios: (1) adversarial injection of private data, and (2) benign attack using an instruction-following benchmark. Our findings highlight the necessity of robust and faithful unlearning methods for safe deployment of language models.

Detecting harmful AI actions is important as AI agents gain adoption. Chain-of-thought (CoT) monitoring is one method widely used to detect adversarial attacks and AI misalignment. However, attackers and misaligned models might evade CoT monitoring through ciphered reasoning: reasoning hidden in encrypted, translated, or compressed text. To assess this risk, we test whether models can perform ciphered reasoning. For each of 28 different ciphers, we fine-tune and prompt up to 10 models to reason in that cipher. We measure model accuracy on math problems as a proxy for reasoning ability. Across the models we test, we find an asymmetry: model accuracy can drop significantly when reasoning in ciphered text, even though models demonstrate comprehension of ciphered text by being able to translate it accurately to English. Even frontier models struggle with lesser-known ciphers, although they can reason accurately in well-known ciphers like rot13. We show that ciphered reasoning capability correlates with cipher prevalence in pretraining data. We also identify scaling laws showing that ciphered reasoning capability improves slowly with additional fine-tuning data. Our work suggests that evading CoT monitoring using ciphered reasoning may be an ineffective tactic for current models and offers guidance on constraining the development of this capability in future frontier models.

Large language models (LLMs) increasingly rely on Chain-of-Thought (CoT) prompting to improve problem-solving and provide seemingly transparent explanations. However, growing evidence shows that CoT often fail to faithfully represent the underlying reasoning process, raising concerns about their reliability in high-risk applications. Although prior studies have focused on mechanism-level analyses showing that CoTs can be unfaithful, they leave open the practical challenge of deciding whether a specific trajectory is faithful to the internal reasoning of the model. To address this gap, we introduce FaithCoT-Bench, a unified benchmark for instance-level CoT unfaithfulness detection. Our framework establishes a rigorous task formulation that formulates unfaithfulness detection as a discriminative decision problem, and provides FINE-CoT (Faithfulness instance evaluation for Chain-of-Thought), an expert-annotated collection of over 1,000 trajectories generated by four representative LLMs across four domains, including more than 300 unfaithful instances with fine-grained causes and step-level evidence. We further conduct a systematic evaluation of eleven representative detection methods spanning counterfactual, logit-based, and LLM-as-judge paradigms, deriving empirical insights that clarify the strengths and weaknesses of existing approaches and reveal the increased challenges of detection in knowledge-intensive domains and with more advanced models. To the best of our knowledge, FaithCoT-Bench establishes the first comprehensive benchmark for instance-level CoT faithfulness, setting a solid basis for future research toward more interpretable and trustworthy reasoning in LLMs.

Identifying inputs that trigger specific behaviours or latent features in language models could have a wide range of safety use cases. We investigate a class of methods capable of generating targeted, linguistically fluent inputs that activate specific latent features or elicit model behaviours. We formalise this approach as context modification and present ContextBench - a benchmark with tasks designed to assess the capabilities of context modification methods across core capabilities and potential safety applications. Our evaluation framework measures both elicitation strength (the degree to which latent features or behaviours are successfully elicited) and linguistic fluency, highlighting how current state-of-the-art methods struggle to balance these objectives. We develop two novel enhancements to Evolutionary Prompt Optimisation, a gradient-based token-editing method: LLM-assistance and diffusion model inpainting, achieving strong performance in balancing elicitation and fluency. We release our benchmark here: https://github.com/lasr-eliciting-contexts/ContextBench.

The fidelity with which neural networks can now generate content such as music presents a scientific opportunity: these systems appear to have learned implicit theories of such content's structure through statistical learning alone. This offers a potentially new lens on theories of human-generated media. When internal representations align with traditional constructs (e.g. chord progressions in music), they show how such categories can emerge from statistical regularities; when they diverge, they expose limits of existing frameworks and patterns we may have overlooked but that nonetheless carry explanatory power. In this paper, focusing on autoregressive music generators, we introduce a method for discovering interpretable concepts using sparse autoencoders (SAEs), extracting interpretable features from the residual stream of a transformer model. We make this approach scalable and evaluable using automated labeling and validation pipelines. Our results reveal both familiar musical concepts and coherent but uncodified patterns lacking clear counterparts in theory or language. As an extension, we show such concepts can be used to steer model generations. Beyond improving model transparency, our work provides an empirical tool for uncovering organizing principles that have eluded traditional methods of analysis and synthesis.

We investigate the mechanistic underpinnings of in-context learning (ICL) in large language models by reconciling two dominant perspectives: the component-level analysis of attention heads and the holistic decomposition of ICL into Task Recognition (TR) and Task Learning (TL). We propose a novel framework based on Task Subspace Logit Attribution (TSLA) to identify attention heads specialized in TR and TL, and demonstrate their distinct yet complementary roles. Through correlation analysis, ablation studies, and input perturbations, we demonstrate that the identified TR and TL heads independently and effectively capture the TR and TL components of ICL. Via steering experiments with a focus on the geometric analysis of hidden states, we reveal that TR heads promote task recognition through aligning hidden states with the task subspace, while TL heads perform rotations to the hidden states within the subspace towards the correct label to facilitate the correct prediction. We also demonstrate how previous findings in various aspects of ICL's mechanism can be reconciled with our attention-head-level analysis of the TR-TL decomposition of ICL, including induction heads, task vectors, and more. Our framework thus provides a unified and interpretable account of how LLMs execute ICL across diverse tasks and settings.

Recently, OpenAI (2025) showed that training against a chain of thought (CoT) monitor can cause obfuscated CoTs, which contain bad behavior the monitor cannot detect. They proposed to keep CoTs monitorable by training only against output monitors that do not have access to CoT. We show that such training can still cause obfuscated CoTs via two mechanisms. First, when a model is trained to produce a safe-looking output, that model may generalize to making its CoTs look safe. Second, since later tokens are conditioned on earlier ones, safe‑looking CoTs may increase the likelihood of safe outputs, causing safe-looking CoTs to be reinforced. We introduce two mitigations to address these two issues, which achieve a Pareto improvement in terms of monitorability and task performance compared to regular training. To our knowledge, we are the first to identify and mitigate these problems. Our work implies that preserving CoT monitorability is more difficult than previously thought; we suggest practical guidelines for AI developers to maintain monitorable CoTs.

We investigate the mechanisms that arise when transformers are trained to solve arithmetic on sequences where tokens are variables whose meaning is determined only through their interactions in-context. While prior work has studied transformers in settings where the answer relies on fixed parametric or geometric information encoded in token embeddings, we devise a new in-context reasoning task where the assignment of tokens to specific algebraic elements varies from one sequence to another. Despite this challenging setup, transformers achieve near-perfect accuracy on the task and even generalize to unseen groups. We develop targeted data distributions to create causal tests of a set of hypothesized mechanisms, and we isolate three mechanisms models consistently learn: commutative copying where a dedicated head copies answers, identity element recognition that distinguishes identity-containing facts, and closure-based cancellation that tracks group membership to constrain valid answers. Our findings show that the kinds of reasoning strategies learned by transformers are dependent on the task structure and that models can develop symbolic reasoning mechanisms when trained to reason in-context about variables whose meanings are not fixed.

Explanations provided by Self-explainable Graph Neural Networks (SE-GNNs) are fundamental for understanding the model's inner workings and for identifying potential misuse of sensitive attributes. Although recent works have highlighted that these explanations can be suboptimal and potentially misleading, a characterization of their failure cases is unavailable. In this work, we identify a critical failure of SE-GNN explanations: explanations can be unambiguously unrelated to how the SE-GNNs infer labels. We show that, on the one hand, many SE-GNNs can achieve optimal true risk while producing these degenerate explanations, and on the other, most faithfulness metrics can fail to identify these failure modes. Our empirical analysis reveals that degenerate explanations can be maliciously planted (allowing an attacker to hide the use of sensitive attributes) and can also emerge naturally, highlighting the need for reliable auditing. To address this, we introduce a novel faithfulness metric that reliably marks degenerate explanations as unfaithful, in both malicious and natural settings. Our code is available on GitHub.

Large reasoning models (LRMs) exhibit unprecedented capabilities in solving complex problems through Chain-of-Thought (CoT) reasoning. However, recent studies reveal that their final answers often contradict their own reasoning traces. We hypothesize that this inconsistency stems from two competing mechanisms for generating answers: CoT reasoning and memory retrieval. To test this hypothesis, we conduct controlled experiments that challenge LRMs with misleading cues during reasoning and/or corrupted answers during retrieval. Our results across models and datasets confirm that both mechanisms operate simultaneously, with their relative dominance influenced by multiple factors: problem domains, model scales, and fine-tuning approaches (e.g., reinforcement learning vs. distillation). The findings reveal a critical limitation in current reasoning fine-tuning paradigms: models can exploit the retrieval mechanism as a shortcut, effectively "hacking" the reward signal and undermining genuine reasoning development. To address this challenge, we introduce FARL, a novel fine-tuning framework that integrates memory unlearning with reinforcement learning. By carefully suppressing retrieval shortcuts during the fine-tuning process, FARL promotes reasoning-dominant behavior and enhances generalizable reasoning capabilities. The code is available at https://github.com/ZJUWYH/FARL.

Retrieval-Augmented Generation (RAG) leverages large language models (LLMs) combined with external contexts to enhance the accuracy and reliability of generated responses. However, reliably attributing generated content to specific context segments, context attribution, remains challenging due to the computationally intensive nature of current methods, which often require extensive fine-tuning or human annotation. In this work, we introduce a novel Jensen–Shannon Divergence driven method to Attribute Response to Context (ARC-JSD), enabling efficient and accurate identification of essential context sentences without additional fine-tuning, gradient-calculation or surrogate modelling. Evaluations on a wide range of RAG benchmarks, such as TyDi QA, Hotpot QA, and Musique, using instruction-tuned LLMs in different scales demonstrate superior accuracy and significant computational efficiency improvements compared to the previous baselines. Furthermore, our mechanistic analysis reveals specific attention heads and multilayer perceptron (MLP) layers responsible for context attribution, providing valuable insights into the internal workings of RAG models and how they affect RAG behaviours.

Compression methods, including quantization, distillation, and pruning, improve the computational efficiency of large reasoning models (LRMs). However, existing studies either fail to sufficiently compare all three compression methods on LRMs or lack in-depth interpretation analysis. In this paper, we investigate how the reasoning capabilities of LRMs are compromised during compression, through performance benchmarking and mechanistic interpretation. To uncover the effects of compression on reasoning performance, we benchmark quantized, distilled, and pruned DeepSeek-R1 models on four reasoning datasets (AIME 2024, FOLIO, Temporal Sequences, and MuSiQue). To precisely locate compression effects on model weights, we adapt difference of means and attribution patching techniques, focusing on the activation of every linear component in compressed LRMs, to interpret fine-grained causal relationships between weights and various reasoning capabilities. This fine-grained interpretation addresses a fundamental question of compression: which weights are the most important for reasoning? Overall, we find dynamically quantized 2.51-bit R1 reaches close-to-R1 performance. With empirical verification, we present three main findings that generalize across both R1 and non-R1 LRMs: (1) Weight count has a greater impact on LRMs' knowledge memorization than reasoning, highlighting the risks of pruning and distillation; (2) The MLP up projection in the final layer of distilled LRMs is one of the most important components, offering a new perspective on locating critical weights - a fundamental problem in model compression; and (3) Current quantization methods overly compress the final-layer modules and MLP gate projections, so protecting just 2% of all weights that are excessively compressed can raise average accuracy by 6.57%, greatly surpassing the state-of-the-art.

Small influential data subsets can dramatically impact model conclusions, with a few data points overturning key findings. While recent work identifies these most influential sets, there is no formal way to tell when maximum influence is excessive rather than expected under natural random sampling variation. We address this gap by developing a principled framework for most influential sets. Focusing on linear least-squares, we derive a convenient exact influence formula and identify the extreme value distributions of maximal influence – the heavy-tailed Fréchet for constant-size sets and heavy-tailed data, and the well-behaved Gumbel for growing sets or light tails. This allows us to conduct rigorous hypothesis tests for excessive influence. We demonstrate through applications across economics, biology, and machine learning benchmarks, resolving contested findings and replacing ad-hoc heuristics with rigorous inference.

Recent advances in multi-modal large reasoning models (MLRMs) have shown significant ability to interpret complex visual content. While these models possess impressive reasoning capabilities, they also introduce novel and underexplored privacy risks. In this paper, we identify a novel category of privacy leakage in MLRMs: Adversaries can infer sensitive geolocation information, such as users' home addresses or neighborhoods, from user-generated images, including selfies captured in private settings. To formalize and evaluate these risks, we propose a three-level privacy risk framework that categorizes image based on contextual sensitivity and potential for geolocation inference. We further introduce DoxBench, a curated dataset of 500 real-world images reflecting diverse privacy scenarios divided into 6 categories. Our evaluation across 13 advanced MLRMs and MLLMs demonstrates that most of these models outperform non-expert humans in geolocation inference and can effectively leak location-related private information. This significantly lowers the barrier for adversaries to obtain users' sensitive geolocation information. We further analyze and identify two primary factors contributing to this vulnerability: (1) MLRMs exhibit strong geolocation reasoning capabilities by leveraging visual clues in combination with their internal world knowledge; and (2) MLRMs frequently rely on privacy-related visual clues for inference without any built-in mechanisms to suppress or avoid such usage. To better understand and demonstrate real-world attack feasibility, we propose GeoMiner, a collaborative attack framework that decomposes the prediction process into two stages consisting of clue extraction and reasoning to improve geolocation performance. Our findings highlight the urgent need to reassess inference-time privacy risks in MLRMs to better protect users' sensitive information.

With the growing importance of trustworthy AI, algorithmic fairness has emerged as a critical concern. Among various fairness notions, group fairness - which measures the model bias between sensitive groups - has received significant attention. While many group-fair models have focused on satisfying group fairness constraints, model uncertainty has received relatively little attention, despite its importance for robust and trustworthy decision-making. To address this, we adopt a Bayesian framework to capture model uncertainty in fair model training. We first define group-fair posterior distributions and then introduce a fair variational Bayesian inference. Then we propose a novel distribution termed matched Gibbs posterior, as a proxy distribution for the fair variational Bayesian inference by employing a new group fairness measure, the matched deviation. A notable feature of matched Gibbs posterior is that it approximates the posterior distribution well under the fairness constraint without requiring heavy computation. Theoretically, we show that the matched deviation has a strong relation to existing group fairness measures, highlighting desirable fairness guarantees. Computationally, by treating the matching function in the matched deviation as a learnable parameter, we develop an efficient MCMC algorithm. Experiments on real-world datasets demonstrates that matched Gibbs posterior outperforms other methods in balancing uncertainty–fairness and utility–fairness trade-offs, while also offering additional desirable properties.

Decision tree ensembles are widely used in critical domains, making robustness and sensitivity analysis essential to their trustworthiness. We study the feature sensitivity problem, which asks whether an ensemble is ``sensitive" to a specified subset of features - such as protected attributes- whose manipulation can alter model predictions. Existing approaches often yield examples of sensitivity that lie far from the training distribution, limiting their interpretability and practical value. We propose a data-aware sensitivity framework that constrains the sensitive examples to remain close to the dataset, thereby producing realistic and interpretable evidence of model weaknesses. To this end, we develop novel techniques for data-aware search using a combination of mixed-integer linear programming (MILP) and satisfibility modulo theories (SMT) encodings. Our contributions are fourfold. Firstly, we strengthen the NP-hardness result for sensitivity verification, showing it holds even for trees of depth 1. Secondly, we develop MILP-optimizations that significantly speed up sensitivity verification for single ensembles and for the first time can also handle multiclass tree ensembles. Thirdly we introduce a data-aware framework generating realistic examples near the training distribution. Finally, we conduct an extensive experimental evaluation on large tree ensembles, demonstrating scalability to ensembles with up to 800 trees of depth 8, achieving substantial improvements over the state of the art. This framework provides a practical foundation for analyzing the reliability and fairness of tree-based models in high-stakes applications.

Multimodal Large Language Models (MLLMs) have shown great promise but require substantial computational resources during inference. Attackers can exploit this by inducing excessive output, leading to resource exhaustion and service degradation. Prior energy-latency attacks aim to increase generation time by broadly shifting the output token distribution away from the EOS token, but they neglect the influence of token-level Part-of-Speech (POS) characteristics on EOS and sentence-level structural patterns on output counts, limiting their efficacy. To address this, we propose \textbf{LingoLoop}, an attack designed to induce MLLMs to generate excessively verbose and repetitive sequences. First, we find that the POS tag of a token strongly affects the likelihood of generating an EOS token. Based on this insight, we propose a \textbf{POS-Aware Delay Mechanism} to postpone EOS token generation by adjusting attention weights guided by POS information. Second, we identify that constraining output diversity to induce repetitive loops is effective for sustained generation. We introduce a \textbf{Generative Path Pruning Mechanism} that limits the magnitude of hidden states, encouraging the model to produce persistent loops. Extensive experiments on models like Qwen2.5-VL-3B demonstrate LingoLoop's powerful ability to trap them in generative loops; it consistently drives them to their generation limits and, when those limits are relaxed, can induce outputs with up to \textbf{367$\times$} more tokens than clean inputs, triggering a commensurate surge in energy consumption. These findings expose significant MLLMs' vulnerabilities, posing challenges for their reliable deployment.

Along with the rapid advancement of numerous Text-to-Video (T2V) models, growing concerns have emerged regarding their safety risks. While recent studies have explored vulnerabilities in models like LLMs, VLMs, and Text-to-Image (T2I) models through jailbreak attacks, T2V models remain largely unexplored, leaving a significant safety gap. To address this gap, we introduce SceneSplit, a novel black-box jailbreak method that works by fragmenting a harmful narrative into multiple scenes, each individually benign. This approach manipulates the generative output space, the abstract set of all potential video outputs for a given prompt, using the combination of scenes as a powerful constraint to guide the final outcome. While each scene individually corresponds to a wide and safe space where most outcomes are benign, their sequential combination collectively restricts this space, narrowing it to an unsafe region and significantly increasing the likelihood of generating a harmful video. This core mechanism is further enhanced through iterative scene manipulation, which bypasses the safety filter within this constrained unsafe region. Additionally, a strategy library that reuses successful attack patterns further improves the attack's overall effectiveness and robustness. To validate our method, we evaluate SceneSplit across 11 safety categories from T2VSafetyBench on T2V models. Our results show that it achieves a high average Attack Success Rate (ASR) of 77.2% on Luma Ray2, 84.1% on Hailuo, 78.2% on Veo2, 78.6% on Kling V1.0, and 68.6% on Sora2, significantly outperforming the existing baselines. Through this work, we demonstrate that current T2V safety mechanisms are vulnerable to attacks that exploit narrative structure, providing new insights for understanding and improving the safety of T2V models.

Current safety evaluations of language models rely on benchmark-based assessments that may miss targeted vulnerabilities. We present RepIt, a simple and data-efficient framework for isolating concept-specific representations in LM activations. While existing steering methods already achieve high attack success rates through broad interventions, RepIt enables a more concerning capability: selective suppression of refusal on targeted concepts while preserving refusal elsewhere. Across five frontier LMs, RepIt produces evaluation-evading models that answer questions related to weapons of mass destruction while still scoring as safe on standard benchmarks. We find the edit of the steering vector localizes to just 100-200 coordinates, and robust concept vectors can be extracted from as few as a dozen examples on a single A6000, highlighting how targeted, hard-to-detect modifications can exploit evaluation blind spots with minimal resources. By demonstrating precise concept disentanglement, this work exposes critical vulnerabilities in current safety evaluation practices and demonstrates an immediate need for more comprehensive, representation-aware assessments.

Visual manipulation localization (VML) aims to identify tampered regions in images and videos, a task that has become increasingly challenging with the rise of advanced editing tools. Existing methods face two central issues. The first is resolution diversity. Resizing or padding can distort subtle forensic cues and introduce unnecessary computational cost. The second is the difficulty of extending spatial models for images to spatio-temporal inputs in videos, which often results in maintaining separate architectures for the two data types. To address these challenges, we propose RelayFormer, a unified framework that adapts to varying resolutions and naturally handles both static and temporal visual data. RelayFormer partitions inputs into fixed-size sub-images and introduces Global Local Relay (GLR) tokens that propagate structured context through a relay-based attention mechanism. This design enables efficient exchange of global cues, such as semantic or temporal consistency, while preserving fine-grained manipulation artifacts. Unlike prior approaches that depend on uniform resizing or sparse attention, RelayFormer scales to variable resolutions and video sequences with minimal overhead. Experiments across diverse benchmarks demonstrate superior performance and strong efficiency, combining resolution adaptivity without interpolation or excessive padding, unified processing for images and videos, and a favorable balance between accuracy and computational cost. Code is available at~\href{https://github.com/WenOOI/RelayFormer}{https://github.com/WenOOI/RelayFormer}.

Despite the growing interest in jailbreaks as an effective red-teaming tool for building safe and responsible large language models (LLMs), flawed evaluation system designs have led to significant discrepancies in their effectiveness assessments. With a systematic measurement study based on 37 jailbreak studies since 2022, we find that existing evaluation systems lack case-specific criteria, resulting in misleading conclusions about their effectiveness and safety implications. In this paper, we introduce GuidedBench, a novel benchmark comprising a curated harmful question dataset and GuidedEval, an evaluation system integrated with detailed case-by-case evaluation guidelines. Experiments demonstrate that GuidedBench offers more accurate evaluations of jailbreak performance, enabling meaningful comparisons across methods. GuidedEval reduces inter-evaluator variance by at least 76.03%, ensuring reliable and reproducible evaluations. We reveal why existing jailbreak benchmarks fail to evaluate accurately and suggest better evaluation practices.

Large Language Models have shown impressive generative capabilities across diverse tasks, but their safety remains a critical concern. Existing post-training alignment methods, such as SFT and RLHF, reduce harmful outputs yet leave LLMs vulnerable to jailbreak attacks, especially advanced optimization-based ones. Recent system-2 approaches enhance safety by adding inference-time reasoning, where models assess potential risks before producing responses. However, we find these methods fail against powerful out-of-distribution jailbreaks, such as AutoDAN-Turbo and Adversarial Reasoning, which conceal malicious goals behind seemingly benign prompts. We observe that all jailbreaks ultimately aim to embed a core malicious intent, suggesting that extracting this intent is key to defense. To this end, we propose ARMOR, which introduces a structured three-step reasoning pipeline: (1) analyze jailbreak strategies from an external, updatable strategy library, (2) extract the core intent, and (3) apply policy-based safety verification. We further develop ARMOR-Think, which decouples safety reasoning from general reasoning to improve both robustness and utility. Evaluations on advanced optimization-based jailbreaks and safety benchmarks show that ARMOR achieves state-of-the-art safety performance, with an average harmful rate of 0.002 and an attack success rate of 0.06 against advanced optimization-based jailbreaks, far below other reasoning-based models. Moreover, ARMOR demonstrates strong generalization to unseen jailbreak strategies, reducing their success rate to zero. These highlight ARMOR’s effectiveness in defending against OOD jailbreak attacks, offering a practical path toward secure and reliable LLMs.

When training large-scale models, the performance typically scales with the number of parameters and the dataset size according to a slow power law. A fundamental theoretical and practical question is whether comparable performance can be achieved with significantly smaller models and substantially less data. In this work, we provide a positive and constructive answer. We prove that a generic permutation-invariant function of $d$ objects can be asymptotically compressed into a function of $\operatorname{polylog} d$ objects with vanishing error, which is proved to be the optimal compression rate. This theorem yields two key implications: (Ia) a large neural network can be compressed to polylogarithmic width while preserving its learning dynamics; (Ib) a large dataset can be compressed to polylogarithmic size while leaving the loss landscape of the corresponding model unchanged. Implication (Ia) directly establishes a proof of the dynamical lottery ticket hypothesis, which states that any ordinary network can be strongly compressed such that the learning dynamics and result remain unchanged. (Ib) shows that a neural scaling law of the form $L\sim d^{-\alpha}$ can be boosted to an arbitrarily fast power law decay, and ultimately to $\exp(-\alpha' \sqrt[m]{d})$.

We use the notion of oracle machines and reductions from computability theory to formalise different Human-in-the-loop (HITL) setups for AI systems, distinguishing between trivial human monitoring (i.e., total functions), single endpoint human action (i.e., many-one reductions), and highly involved human-AI interaction (i.e., Turing reductions). We then proceed to show that the legal status and safety of different setups vary greatly. We present a taxonomy to categorise HITL failure modes, highlighting the practical limitations of HITL setups. We then identify omissions in UK and EU legal frameworks, which focus on HITL setups that may not always achieve the desired ethical, legal, and sociotechnical outcomes. We suggest areas where the law should recognise the effectiveness of different HITL setups and assign responsibility in these contexts, avoiding human `scapegoating'. Our work shows an unavoidable trade-off between attribution of legal responsibility, and technical explainability. Overall, we show how HITL setups involve many technical design decisions, and can be prone to failures out of the humans' control. Our formalisation and taxonomy opens up a new analytic perspective on the challenges in creating HITL setups, helping inform AI developers and lawmakers on designing HITL setups to better achieve their desired outcomes.

Machine learning models trained on real-world data often inherit and amplify biases against certain social groups, raising urgent concerns about their deployment at scale. While numerous bias mitigation methods have been proposed, comparing the effectiveness of bias mitigation methods remains difficult due to heterogeneous datasets, inconsistent fairness metrics, isolated evaluation of vision versus multi-modal models, and insufficient hyperparameter tuning that undermines fair comparisons. We introduce NH-Fair, a unified benchmark for fairness without harm that spans both vision models and large vision–language models (LVLMs) under standardized data, metrics, and training protocols, covering supervised and zero-shot regimes. Our key contributions are: (1) a systematic ERM tuning study that identifies training choices with large influence on both utility and disparities, yielding empirically grounded guidelines to help practitioners reduce expensive hyperparameter tuning space in achieving strong fairness and accuracy; (2) evidence that many debiasing methods do not reliably outperform a well-tuned ERM baseline, whereas a composite data-augmentation method consistently delivers parity gains without sacrificing utility, emerging as a promising practical strategy. (3) an analysis showing that while LVLMs achieve higher average accuracy, they still exhibit subgroup disparities, and gains from scaling are typically smaller than those from architectural or training-protocol choices. NH-Fair provides a reproducible, tuning-aware pipeline for rigorous, harm-aware fairness evaluation.

Geo-distributed training and Federated Learning (FL) provide viable solutions to address the substantial data and computational resource needs associated with training large language models (LLMs). However, we empirically demonstrate that a single adversarial participant can significantly compromise the safety alignment of LLMs through malicious training, exposing serious security risks. We identify two existing server-side defense strategies that effectively counter naive jailbreak attacks—Task Performance Check (TPC), which filters out model updates with low downstream performance, and Malicious Output Scrutiny (MOS), which detects harmful outputs by prompting uploaded models with malicious queries. To evade both defenses, we design a trigger-based jailbreak variant that preserves downstream performance using a novel regularization method to limit the excessive model updates on jailbreak datasets. We further conceal malicious triggers by mixing the malicious dataset with pseudo-contrastive safety-aligned answers to maintain the original safety alignment. Experiments on three widely-used safety-aligned LLMs show that a single adversarial participant can implant triggers into the global model without degrading downstream performance, achieving an 80\% attack success rate (ASR) with a 7\% low detection true rate (DTR).

Large language models (LLMs) generate increasingly human-like text, raising concerns about misinformation and authenticity. Detecting AI-generated text remains challenging: existing methods often underperform, especially on short texts, require probability access unavailable in real-world black-box settings, incur high costs from multiple calls, or fail to generalize across models. We propose Disrupt-and-Recover (D&amp;R), a recovery-based detection framework grounded in posterior concentration. D&amp;R disrupts text via model-free Within-Chunk Shuffling, performs a single black-box LLM recovery, and measures semantic–structural recovery similarity as a proxy for concentration. This design ensures efficiency, black-box practicality, and is theoretically supported under the concentration assumption. Extensive experiments across four datasets and six source models show that D&amp;R achieves state-of-the-art performance, with AUROC 0.96 on long texts and 0.87 on short texts, surpassing the strongest baseline by +0.08 and +0.14. D&amp;R further remains robust under source–recovery mismatch and model variation. Our code and data is available at https://github.com/Yuxia-Sun/D-R.

Advances in Large Language Models (LLMs) have enabled a new class of \textbf{\textit{self-evolving agents}} that autonomously improve through environmental interaction, demonstrating strong capabilities. However, self-evolution also introduces novel risks overlooked by current safety research. In this work, we study case where an agent's self-evolution deviates in unintended ways, leading to undesirable or even harmful outcomes. We refer to this as \textit{\textbf{Misevolution}}. We evaluate misevolution along four key evolutionary pathways: model, memory, tool, and workflow. Our empirical findings reveal that misevolution is a widespread risk, affecting agents built even on top-tier LLMs (\textit{e.g.}, Gemini-2.5-Pro). Different emergent risks are observed, such as degradation of safety alignment after memory accumulation, or unintended introduction of vulnerabilities in tool creation and reuse. To our knowledge, this is the first study to systematically conceptualize misevolution and provide empirical evidence of its occurrence, highlighting an urgent need for new safety paradigms for self-evolving agents. Finally, we discuss potential mitigation strategies to inspire further research on building safer and more trustworthy self-evolving agents.

Data contamination poses a significant threat to the reliable evaluation of Large Language Models (LLMs). This issue arises when benchmark samples may inadvertently appear in training sets, compromising the validity of reported performance. While detection methods have been developed for the pre-training and Supervised Fine-Tuning stages, a critical research gap exists for the increasingly significant phase of Reinforcement Learning (RL) post-training. As RL post-training becomes pivotal for advancing LLM reasoning, the absence of specialized contamination detection methods in this paradigm presents a critical vulnerability. To address this, we conduct the first systematic study of data detection within RL post-training scenario and propose Self-Critique. Our method is motivated by a key observation: after RL phase, the output entropy distribution of LLMs tends to collapse into highly specific and sparse modes. Self-Critique probes for the underlying policy collapse, i.e., the model's convergence to a narrow reasoning path, which causes this entropy reduction. To facilitate this research, we also introduce RL-MIA, a benchmark constructed to simulate this specific contamination scenario. Extensive experiments show that Self-Critique significantly outperforms baseline methods across multiple models and contamination tasks, achieving an AUC improvement of up to 30%. Whereas existing methods are close to a random guess for RL-phase contamination, our method makes detection possible.

Conventional preference learning methods often prioritize opinions held more widely when aggregating preferences from multiple evaluators. This may result in policies that are biased in favor of some types of opinions or groups and susceptible to strategic manipulation. To address this issue, we develop a novel preference learning framework capable of aligning aggregate opinions and policies proportionally with the true population distribution of evaluator preferences. Grounded in social choice theory, our approach infers the feasible set of evaluator population distributions directly from pairwise comparison data. Using these estimates, the algorithm constructs a policy that satisfies foundational axioms from social choice theory, namely monotonicity and Pareto efficiency, as well as our newly-introduced axioms of population-proportional alignment and population-bounded manipulability. Moreover, we propose a soft-max relaxation method that smoothly trades off population-proportional alignment with the selection of the Condorcet winner (which beats all other options in pairwise comparisons). Finally, we validate the effectiveness and scalability of our approach through experiments on both tabular recommendation tasks and large language model alignment.

Large Reasoning Models (LRMs) with long chain-of-thought capabilities, optimized via reinforcement learning with verifiable rewards (RLVR), excel at objective reasoning tasks like mathematical problem solving and code generation. However, RLVR is known for degrading generation diversity, which causes LRMs to fall short on subjective reasoning that has multiple answers depending on different role perspectives. While recent studies recognize the importance of diversity-enhanced training in objective reasoning, limited attention has been given to subjective tasks. In this paper, we find that subjective reasoning can be improved by introducing perspective diversity and token-level diversity, with the former one providing a coherent scaffolding anchored to a real-world stakeholder group and the latter one broadening the answer search space. We propose MultiRole-R1, a diversity-enhanced training framework featuring an unsupervised data construction pipeline that synthesizes reasoning chains incorporating various role perspectives. It also employs reinforcement learning via Group Relative Policy Optimization with reward shaping, taking diversity as a reward signal in addition to verifiable reward. Training on subjective tasks solely, MultiRole-R1 increases the in-domain and out-of-domain accuracy by 14.1% and 7.64%, and even enhances the performance on advanced math reasoning such as AIME 2024. We further show that diversity is a more consistent indicator of accuracy than reasoning length.

Algorithmic fairness is a socially crucial topic in real-world applications of AI. Among many notions of fairness, subgroup fairness is widely studied when multiple sensitive attributes (e.g., gender, race, and age) are present. However, as the number of sensitive attributes grows, the number of subgroups increases accordingly, creating heavy computational burden and data sparsity problem (i.e., subgroups with very small sample sizes). In this paper, we develop a novel learning algorithm for subgroup fairness that resolves these issues by focusing on sufficiently large subgroups as well as marginal fairness (fairness for each sensitive attribute). To this end, we formalize a notion of subgroup-subset fairness and introduce a corresponding distributional fairness measure called the supremum Integral Probability Metric (supIPM). Building on this formulation, we propose the Doubly Regressing Adversarial learning for subgroup Fairness (DRAF) algorithm, which reduces a surrogate fairness gap for supIPM with much less computation than directly reducing supIPM. Theoretically, we prove that the proposed surrogate fairness gap is an upper bound of supIPM. Empirically, we show that the DRAF algorithm outperforms baseline methods on benchmark datasets, particularly when the number of sensitive attributes is large so that many subgroups are very small.

Large Language Models (LLMs) exhibit strong but shallow alignment: they directly refuse harmful queries when a refusal is expected at the very start of an assistant turn, yet this protection collapses once a harmful continuation is underway (either through the adversarial attacks or via harmful assistant-prefill attacks). This raises a fundamental question: Can the innate shallow alignment in LLMs be unlocked to ensure safety at arbitrary generation depths? To achieve this goal, we propose Any-Depth Alignment (ADA) an effective inference-time defense with negligible overhead. ADA is built based on our observation that alignment is concentrated in the assistant header tokens through repeated use in shallow-refusal training, and these tokens possess the model’s strong alignment priors. By reintroducing these tokens mid-stream, ADA induces the model to reassess harmfulness and recover refusals at any point in generation. Across diverse open-source model families (Llama, Gemma, Mistral, Qwen, DeepSeek, and gpt-oss), ADA achieves robust safety performance without requiring any changes to the base model's parameters. It secures a near-100% refusal rate against challenging adversarial prefill attacks ranging from dozens to thousands of tokens. Furthermore, ADA reduces the average success rate of prominent adversarial prompt attacks (such as GCG, AutoDAN, PAIR, and TAP) to below 3%. This is all accomplished while preserving benign utility with minimal over-refusal and maintaining resilience even after the base model undergoes subsequent instruction tuning.

Large language models (LLMs) often encode social biases tied to sensitive features such as race and gender, undermining fairness in downstream tasks even after instruction tuning. Conventional debiasing methods require expensive fine-tuning, are tied to specific architectures, or operate only at the input or decoding stage while neglecting attention-level representations, which can result in compromised task performance. Moreover, most approaches are tailored to single-attribute settings and do not explicitly address scenarios with multiple, overlapping protected attributes and their intersections. This paper proposes a novel method of multi-feature quantized attention regularization (MQAR) to mitigate multi-feature bias by injecting a structured quantization into frozen self-attention layers. MQAR disentangles attribute-specific activations through vector-quantized regularization and uses a discriminator-guided autoencoding regularizer to adversarially suppress protected-attribute information while preserving task-relevant semantics. Crucially, the proposed method operates without modifying the backbone parameters or accessing pre-training data, ensuring architecture-agnostic applicability and minimizing representation distortion. MQAR is evaluated on five diverse LLMs (BERT, T5, GPT-Neo, Mixtral, and LLaMA 3.2) using three standard bias benchmarks (WinoBias, StereoSet, and CrowS-Pairs). Across these models, MQAR consistently reduces bias for multiple protected attributes and their intersections while maintaining downstream accuracy within at most 0.4 \%, on average, of non-debiased baselines on sentiment analysis, abusive language detection, and text generation tasks. These findings highlight quantized attention regularization as a scalable and effective method for mitigating social bias in modern language models.

We propose DIVERSE, a framework for systematically exploring the Rashomon set of deep neural networks, the collection of models that match a reference model’s accuracy while differing in their predictive behavior. DIVERSE augments a pretrained model with Feature-wise Linear Modulation (FiLM) layers and uses Covariance Matrix Adaptation Evolution Strategy (CMA-ES) to search a latent modulation space, generating diverse model variants without retraining or gradient access. Across MNIST, PneumoniaMNIST, and CIFAR-10, DIVERSE uncovers multiple high-performing yet functionally distinct models. Our experiments show that DIVERSE offers a competitive and efficient exploration of the Rashomon set, making it feasible to construct diverse sets that maintain robustness and performance while supporting well-balanced model multiplicity. While retraining remains the baseline to generate Rashomon sets, DIVERSE achieves comparable diversity at reduced computational cost.

Online learning in arbitrary, and possibly adversarial, environments has been extensively studied in sequential decision-making, and it is closely connected to equilibrium computation in game theory. Most existing online learning algorithms rely on \emph{numeric} utility feedback from the environment, which may be unavailable in human-in-the-loop applications and/or may be restricted by privacy concerns. In this paper, we study an online learning model in which the learner only observes a \emph{ranking} over a set of proposed actions at each timestep. We consider two ranking mechanisms: rankings induced by the \emph{instantaneous} utility at the current timestep, and rankings induced by the \emph{time-average} utility up to the current timestep, under both \emph{full-information} and \emph{bandit} feedback settings. Using the standard external-regret metric, we show that sublinear regret is impossible with instantaneous-utility ranking feedback in general. Moreover, when the ranking model is relatively deterministic, \emph{i.e.}, under the Plackett-Luce model with a temperature that is sufficiently small, sublinear regret is also impossible with time-average utility ranking feedback. We then develop new algorithms that achieve sublinear regret under the additional assumption that the utility sequence has sublinear total variation. Notably, for full-information time-average utility ranking feedback, this additional assumption can be removed. As a consequence, when all players in a normal-form game follow our algorithms, repeated play yields an approximate coarse correlated equilibrium. We also demonstrate the effectiveness of our algorithms in an online large-language-model routing task.

The proliferation of realistic deepfakes has driven the development of numerous benchmark datasets to support detection research. Despite their increasing volume and diversity, no prior effort has systematically consolidated these resources into a unified framework for large-scale model training, nor has there been a massively pre-trained model tailored to deepfake detection. In this work, we introduce Multi-modal Multi-type Integrated Deepfake Dataset (MMI-DD), a large-scale resource containing 3.6 million facial images, the largest collection to date. It unifies diverse benchmarks with uniform preprocessing, and further provides fine-grained annotations across four deepfake types, as well as VLM-generated descriptions capturing both facial and environmental attributes for each image. By leveraging this comprehensive multi-modal dataset, we construct a foundational deepfake knowledge space that empowers our model to discern a broad spectrum of synthetic media. Our method, $SD^2$ (Scalable Deepfake Detection), refines CLIP for deepfake detection, optimizing image-text classification with rich, type-specific labels. We enhance this with intermediate visual features capturing low-level cues and text label separation loss for stability. We further leverage VLM-generated descriptions and contrastive learning to expand the scope of forgery knowledge, reducing overfitting and enhancing generalization. Extensive experiments on challenging deepfake datasets and AIGC benchmark demonstrate the effectiveness, scalability, and real-world applicability of our approach.

Large Language Models (LLMs) rely on safety alignment to produce socially acceptable responses. However, this behavior is known to be brittle: further fine-tuning, even on benign or lightly contaminated data, can degrade safety and reintroduce harmful behaviors. A growing body of work suggests that alignment may correspond to identifiable directions in weight space, forming subspaces that could, in principle, be isolated or preserved to defend against misalignment. In this work, we conduct a comprehensive empirical study of this perspective. We examine whether safety-relevant behavior is concentrated in specific linear subspaces, whether it can be separated from general-purpose learning, and whether harmfulness arises from distinguishable patterns in activations. Across both weight and activation spaces, our findings are consistent: subspaces that amplify safe behaviors also amplify useful ones, and prompts with different safety implications activate overlapping representations. Rather than residing in distinct directions, we show that safety is highly entangled with the general learning components of the model. This suggests that subspace-based defenses face fundamental limitations and underscores the need for alternative strategies to preserve safety under continued training. We corroborate these findings with multiple experiments on five open-source LLMs from the Llama and Qwen families. Our code is publicly available at: https://github.com/CERT-Lab/safety-subspaces.

Deep neural networks often exhibit degraded performance under domain shifts due to reliance on domain-specific features. Existing domain generalization (DG) methods attempt to mitigate this during training but lack mechanisms to adaptively correct domain-specific reliance once it emerges. We propose Identify and Unlearn (IU), a model-agnostic module that continually mitigates such reliance post-epoch. We introduce an unlearning score to identify training samples that disproportionately increase model complexity while contributing little to generalization, and an Inter-Domain Variance (IDV) metric to reliably identify domain-specific channels. To suppress the adverse influence of identified samples, IU employs a Domain-Specific Gradient-Ascent (DSGA) procedure that selectively removes domain-specific features while preserving domain-invariant features. Extensive experiments across seven benchmarks and fifteen DG baselines show that IU consistently improves out-of-distribution generalization, achieving average accuracy gains of up to 3.0\%.

Video understanding has made remarkable progress in recent years, largely driven by advances in deep models and the availability of large-scale annotated datasets. However, the robustness of these models to domain shifts encountered in real-world video applications remains a critical yet underexplored problem, limiting their practical reliability. To address this problem, we introduce \textbf{V}ideo \textbf{U}nderstanding \textbf{D}omain \textbf{G}eneralization (\textbf{VUDG}), the first dataset designed specifically for evaluating domain generalization in video understanding. VUDG contains videos from 11 distinct domains that cover three types of domain shifts, and maintains semantic consistency across different domains to ensure fair and meaningful evaluation. We propose a multi-expert progressive annotation framework to efficiently annotate videos with structured question-answer pairs designed for domain generalization. Extensive experiments on 9 representative Large Vision-Language Models (LVLMs) and several traditional video question answering methods show that most models (including state-of-the-art LVLMs) suffer performance degradation under domain shifts. These results highlight the challenges posed by VUDG and the difference in the robustness of current models to data distribution shifts. We believe VUDG provides a critical resource to benefit future research in domain generalization for video understanding.

Most familiar equilibrium concepts, such as Nash and correlated equilibrium, guarantee only that no single player can improve their utility by deviating unilaterally. They offer no guarantees against profitable coordinated deviations by coalitions. Although the literature proposes notions to address multilateral deviations (\emph{e.g.}, strong Nash and coalition-proof equilibrium), these generally fail to exist. In this paper, we study a solution concept that accommodates multi-player deviations and is guaranteed to exist. We prove a fixed-parameter lower bound on the complexity of computing such an equilibrium and present an algorithm that matches this bound.

Large Language Models (LLMs) reasoning abilities are increasingly being applied to classical board and card games, but the dominant approach---involving prompting for direct move generation---has significant drawbacks. It relies on the model's implicit fragile pattern-matching capabilities, leading to frequent illegal moves and strategically shallow play. Here we introduce an alternative approach: We use the LLM to translate natural language rules and game trajectories into a formal, executable world model represented as Python code. This generated model---comprising functions for state transition, legal move enumeration, and termination checks---serves as a verifiable simulation engine for high-performance planning algorithms like Monte Carlo tree search (MCTS). In addition, we prompt the LLM to generate heuristic value functions (to make MCTS more efficient), and inference functions (to estimate hidden states in imperfect information games). Our method offers three distinct advantages compared to directly using the LLM as a policy: (1) Verifiability: The generated CWM serves as a formal specification of the game's rules, allowing planners to algorithmically enumerate valid actions and avoid illegal moves, contingent on the correctness of the synthesized model; (2) Strategic Depth: We combine LLM semantic understanding with the deep search power of classical planners; and (3) Generalization: We direct the LLM to focus on the meta-task of data-to-code translation, enabling it to adapt to new games more easily. We evaluate our agent on 10 different games, of which 4 are novel and created for this paper. 5 of the games are fully observed (perfect information), and 5 are partially observed (imperfect information). We find that our method outperforms or matches Gemini 2.5 Pro in 9 out of the 10 considered games.

We study the problem of fixed budget pure exploration in reinforcement learning. The goal is to identify a near-optimal policy, given a fixed budget on the number of interactions with the environment. Unlike the standard PAC setting, we do not require the target error level $\epsilon$ and failure rate $\delta$ as input. We propose novel algorithms and provide, to the best of our knowledge, the first instance-dependent $\epsilon$-uniform guarantee, meaning that the probability that $\epsilon$-correctness is ensured can be obtained simultaneously for all $\epsilon$ above a budget-dependent threshold. It characterizes the budget requirements in terms of the problem-specific hardness of exploration. As a core component of our analysis, we derive a $\epsilon$-uniform guarantee for the multiple bandit problem—solving multiple multi-armed bandit instances simultaneously—which may be of independent interest. To enable our analysis, we also develop tools for reward-free exploration under the fixed-budget setting, which we believe will be useful for future work.

Boolean functions constitute a fundamental object of study in machine learning and theoretical computer science. Among their various complexity measures, Fourier sparsity, the number of nonzero coefficients in a function’s Fourier expansion, serves as a natural indicator of structural simplicity. For more than three decades, the problem of learning Boolean functions with sparse Fourier representations has occupied a central place in computational learning theory. A major line of progress has produced algorithms whose complexities depend primarily on the sparsity parameter itself. However, these methods typically assume that this parameter is known in advance. In this work, we explore the problem of Fourier sparsity testing, which naturally relates to this question. Given query access to a Boolean function $f : \mathbb{F}_2^n \to \{ -1, +1 \}$, we seek to determine whether it is $s$-Fourier sparse or far (under Hamming distance) from every such function. Our contributions are twofold. On the algorithmic side, we design a new tester with query complexity $\widetilde{O}(s^4)$, independent of the ambient dimension. On the lower bound side, we prove that any tester requires at least $\Omega(s)$ queries. Both bounds improve upon the best known results of Gopalan et al. (SICOMP 2011), who obtained a tester with query complexity $\widetilde{O}(s^{14})$ and a lower bound of $\Omega(\sqrt{s})$. For the upper bound, we introduce a refined notion of a sampler inspired by the junta testing framework and combine it with $\ell_1$-minimization-based compressed sensing techniques. In doing so, we develop a novel method for sampling leaves of parity decision trees associated with Fourier-sparse Boolean functions. The lower bound is obtained via a reduction from communication complexity, leveraging structural properties of Fourier coefficients of a specific class of cryptographically hard functions.

Transformers have achieved great success across a wide range of applications, yet the theoretical foundations underlying their success remain largely unexplored. To demystify the strong capacities of transformers applied to versatile scenarios and tasks, we theoretically investigate utilizing transformers as students to learn from a class of teacher models. Specifically, the teacher models covered in our analysis include convolution layers with average pooling, graph convolution layers, and various classic statistical learning models, including a variant of sparse token selection models [Sanford et al., 2023, Wang et al., 2024] and group-sparse linear predictors [Zhang et al., 2025]. When learning from this class of teacher models, we prove that one-layer transformers with simplified "position-only'' attention can successfully recover all parameter blocks of the teacher models, thus achieving the optimal population loss. Building upon the efficient mimicry of trained transformers towards teacher models, we further demonstrate that they can generalize well to a broad class of out-of-distribution data under mild assumptions. The key in our analysis is to identify a fundamental bilinear structure shared by various learning tasks, which enables us to establish unified learning guarantees for these tasks when treating them as teachers for transformers.

Uniform sampling is a highly efficient method for data summarization. However, its effectiveness in producing coresets for clustering problems is not yet well understood, primarily because it generally does not yield a strong coreset, which is the prevailing notion in the literature. We formulate \emph{stable coresets}, a notion that is intermediate between the standard notions of weak and strong coresets, and effectively combines the broad applicability of strong coresets with highly efficient constructions, through uniform sampling, of weak coresets. Our main result is that a uniform sample of size $O(\epsilon^{-2}\log d)$ yields, with high constant probability, a stable coreset for $1$-median in $\mathbb{R}^d$ under the $\ell_1$ metric. We then leverage the powerful properties of stable coresets to easily derive new coreset constructions, all through uniform sampling, for $\ell_1$ and related metrics, such as Kendall-tau and Jaccard. We also show applications to fair rank aggregation and to approximation algorithms for $k$-median problem in these metric spaces. Our experiments validate the benefits of stable coresets in practice, in terms of both construction time and approximation quality.

Sharpness (of the loss minima) is widely believed to be a good indicator of generalization of neural networks. Unfortunately, the correlation between existing sharpness measures and the generalization is not that strong as expected, sometimes even contradiction occurs. To address this problem, a key observation in this paper is: what really matters for the generalization is the *average spread* (or unevenness) of the spectrum of loss Hessian $\mathbf{H}$. For this reason, the conventional sharpness measures, such as the trace sharpness $\operatorname{tr}(\mathbf{H})$, which cares about the *average value* of the spectrum, or the max-eigenvalue sharpness $\lambda_{\max}(\mathbf{H})$), which concerns the *maximum spread* of the spectrum, are not sufficient to well predict the generalization. To finely characterize the average spread of the Hessian spectrum, we leverage the notion of *Rényi entropy* in information theory, which is capable of capturing the unevenness of a probability vector and thus can be extended to describe the unevenness for a general non-negative vector (which is the case for the Hessian spectrum at the loss minima). In specific, in this paper we propose the *Rényi sharpness*, which is defined as the negative of the Rényi entropy of loss Hessian $\mathbf{H}$. Extensive experiments demonstrate that Rényi sharpness exhibit *strong* and *consistent* correlation with generalization in various scenarios. Moreover, on the theoretical side, two generalization bounds with respect to the Rényi sharpness are established, by exploiting the desirable reparametrization invariance property of Rényi sharpness. Finally, as an initial attempt to take advantage of the Rényi sharpness for regularization, Rényi Sharpness Aware Minimization (RSAM) algorithm is proposed where a variant of Rényi Sharpness is used as the regularizer. It turns out this RSAM is competitive with the state-of-the-art SAM algorithms, and far better than the conventional SAM algorithm based on the max-eigenvalue sharpness.

We study computationally and statistically efficient reinforcement learning under the linear $Q^{\pi}$ realizability assumption, where any policy's $Q$-function is linear in a given state-action feature representation. Prior methods in this setting are either computationally intractable, or require (local) access to a simulator. In this paper, we propose a computationally efficient online RL algorithm, named *Frozen Policy Iteration*, under the linear $Q^{\pi}$ realizability setting that works for Markov Decision Processes (MDPs) with stochastic initial states, stochastic rewards and deterministic transitions. Our algorithm achieves a regret bound of $\widetilde{O}(\sqrt{d^2H^6T})$, where $d$ is the dimensionality of the feature space, $H$ is the horizon length, and $T$ is the total number of episodes. Our regret bound is optimal for linear (contextual) bandits which is a special case of our setting with $H = 1$. Existing policy iteration algorithms under the same setting heavily rely on repeatedly sampling the same state by access to the simulator, which is not implementable in the online setting with stochastic initial states studied in this paper. In contrast, our new algorithm circumvents this limitation by strategically using only high-confidence part of the trajectory data and freezing the policy for well-explored states, which ensures that all data used by our algorithm remains effectively *on-policy* during the whole course of learning. We further demonstrate the versatility of our approach by extending it to the Uniform-PAC setting and to function classes with bounded eluder dimension.

Knowledge Distillation (KD) is a central paradigm for transferring knowledge from a large teacher network to a typically smaller student model, often by leveraging soft probabilistic outputs. While KD has shown strong empirical success in numerous applications, its theoretical underpinnings remain only partially understood. In this work, we adopt a Bayesian perspective on KD to rigorously analyze the convergence behavior of students trained with Stochastic Gradient Descent (SGD). We study two regimes: $(i)$ when the teacher provides the exact Bayes Class Probabilities (BCPs); and $(ii)$ supervision with noisy approximations of the BCPs. Our analysis shows that learning from BCPs yields variance reduction and removes neighborhood terms in the convergence bounds compared to one-hot supervision. We further characterize how the level of noise affects generalization and accuracy. Motivated by these insights, we advocate the use of Bayesian deep learning models, which typically provide improved estimates of the BCPs, as teachers in KD. Consistent with our analysis, we experimentally demonstrate that students distilled from Bayesian teachers not only achieve higher accuracies (up to +4.27\%), but also exhibit more stable convergence (up to 30\% less noise), compared to students distilled from deterministic teachers.

Neural networks trained with gradient descent often learn solutions of increasing complexity over time, a phenomenon known as simplicity bias. Despite being widely observed across architectures, existing theoretical treatments lack a unifying framework. We present a theoretical framework that explains a simplicity bias arising from saddle-to-saddle learning dynamics for a general class of neural networks, incorporating fully-connected, convolutional, and attention-based architectures. Here, simple means expressible with few hidden units, i.e., hidden neurons, convolutional kernels, or attention heads. Specifically, we show that linear networks learn solutions of increasing rank, ReLU networks learn solutions with an increasing number of kinks, convolutional networks learn solutions with an increasing number of convolutional kernels, and self-attention models learn solutions with an increasing number of attention heads. By analyzing fixed points, invariant manifolds, and dynamics of gradient descent learning, we show that saddle-to-saddle dynamics operates by iteratively evolving near an invariant manifold, approaching a saddle, and switching to another invariant manifold. Our analysis also disentangles data-induced and initialization-induced saddle-to-saddle dynamics. In particular, the former leads to low-rank weights while the latter to sparse weights. Equipped with the theory, we predict the effects of data distribution and weight initialization on the duration and number of plateaus in learning. Overall, our theory offers a framework for understanding when and why gradient descent progressively learns increasingly complex solutions.

In this paper, we provide sufficient conditions of benign overfitting of fixed width leaky ReLU two-layer neural network classifiers trained on mixture data via gradient descent. Our results are derived by establishing directional convergence of the network parameters and classification error bound of the convergent direction. Our classification error bound also lead to the discovery of a newly identified phase transition. Previously, directional convergence in (leaky) ReLU neural networks was established only for gradient flow. Due to the lack of directional convergence, previous results on benign overfitting were limited to those trained on nearly orthogonal data. All of our results hold on mixture data, which is a broader data setting than the nearly orthogonal data setting in prior work. We demonstrate our findings by showing that benign overfitting occurs with high probability in a much wider range of scenarios than previously known. Our results also allow us to characterize cases when benign overfitting provably fails even if directional convergence occurs. Our work thus provides a more complete picture of benign overfitting in leaky ReLU two-layer neural networks.

Dense associative memory, a fundamental instance of modern Hopfield networks, can store a large number of memory patterns as equilibrium states of recurrent networks. While the stationary-state storage capacity has been investigated, its dynamical properties have not yet been discussed. In this paper, we analyze the dynamics using an exact approach based on generating functional analysis. We show results on convergence properties of memory retrieval, such as the convergence time and the size of the attraction basins. Our analysis enables a quantitative evaluation of the convergence time and the storage capacity of dense associative memory, which is useful for model design. Unlike the traditional Hopfield model, the retrieval of a pattern does not act as additional noise to itself, suggesting that the structure of modern networks makes recall more robust. Furthermore, the methodology addressed here can be applied to other energy-based models, and thus has the potential to contribute to the design of future architectures.

We study the prophet inequality, a fundamental problem in online decision-making and optimal stopping, in a practical setting where rewards are observed only through noisy realizations and reward distributions are unknown. At each stage, the decision-maker receives a noisy reward whose true value follows a linear model with an unknown latent parameter, and observes a feature vector drawn from a distribution. To address this challenge, we propose algorithms that integrate learning and decision-making via lower-confidence-bound (LCB) thresholding. In the i.i.d. setting, we establish that both an Explore-then-Decide strategy and an $\varepsilon$-Greedy variant achieve the sharp competitive ratio of $1 - 1/e$, under a mild condition on the optimal value. For non-identical distributions, we show that a competitive ratio of $1/2$ can be guaranteed against a relaxed benchmark. Moreover, with limited window access to past rewards, the tight ratio of $1/2$ against the optimal benchmark is achieved.

Matrix denoising is a crucial component in machine learning, offering valuable insights into the behavior of learning algorithms (Bishop and Nasrabadi, 2006). This paper focuses on the rectangular matrix denoising problem, which involves estimating the left and right singular vectors of a rank-one matrix that is corrupted by additive noise. Traditional algorithms for this problem often exhibit high computational complexity, leading to the widespread use of gradient descent (GD)-based estimation methods with a quadratic cost function. However, the learning dynamics of these GD-based methods, particularly the analytical solutions that describe their exact trajectories, have been largely overlooked in existing literature. To fill this gap, we investigate the learning dynamics in detail, providing convergence proofs and asymptotic analysis. By leveraging tools from large random matrix theory, we derive a closed-form solution for the learning dynamics, characterized by the inner products of the estimates and the ground truth vectors. We rigorously prove the almost sure convergence of these dynamics as the signal dimensions tend to infinity. Additionally, we analyze the asymptotic behavior of the learning dynamics in the large-time limit, which aligns with the well-known Baik-Ben Arous-Péchée phase transition phenomenon n (Baik et al., 2005). Experimental results support our theoretical findings, demonstrating that when the signal-to-noise ratio (SNR) surpasses a critical threshold, learning converges rapidly from an initial value close to the stationary point. In contrast, estimation becomes infeasible when the ratio of the inner products between the initial left and right vectors and their corresponding ground truth vectors reaches a specific value, which depends on both the SNR and the data dimensions.

Multi-agent learning faces a fundamental tension: leveraging distributed collaboration without sacrificing the personalization needed for diverse agents. This tension intensifies when aiming for full personalization while adapting to unknown heterogeneity levels—gaining collaborative speedup when agents are similar, without performance degradation when they are different. Embracing the challenge, we propose personalized collaborative learning (PCL), a novel framework for heterogeneous agents to collaboratively learn personalized solutions with seamless adaptivity. Through carefully designed bias correction and importance correction mechanisms, our method AffPCL robustly handles both environment and objective heterogeneity. We prove that AffPCL reduces sample complexity over independent learning by a factor of $\max\\{n^{-1}, \delta\\}$, where $n$ is the number of agents and $\delta\in[0,1]$ measures their heterogeneity. This *affinity-based* acceleration automatically interpolates between the linear speedup of federated learning in homogeneous settings and the baseline of independent learning, without requiring prior knowledge of the system. Our analysis further reveals that an agent may obtain linear speedup even by collaborating with arbitrarily dissimilar agents, unveiling new insights into personalization and collaboration in the high heterogeneity regime.

Spectral bias, the tendency of neural networks to learn low frequencies first, can be both a blessing and a curse. While it enhances the generalization capabilities by suppressing high-frequency noise, it can be a limitation in scientific tasks that require capturing fine-scale structures. The delayed generalization phenomenon known as grokking is another barrier to rapid training of neural networks. Grokking has been hypothesized to arise as learning transitions from the NTK to the feature-rich regime. This paper explores the impact of preconditioned gradient descent (PGD), such as Gauss-Newton, on spectral bias and grokking phenomena. We demonstrate through theoretical and empirical results how PGD can mitigate issues associated with spectral bias. Additionally, building on the rich learning grokking hypothesis, we study how PGD can be used to reduce delays associated with grokking. Our conjecture is that PGD, without the impediment of spectral bias, enables uniform exploration of the parameter space in the NTK regime. Our experimental results confirm this prediction, providing strong evidence that grokking represents a transitional behavior between the lazy regime characterized by the NTK and the rich regime. These findings deepen our understanding of the interplay between optimization dynamics, spectral bias, and the phases of neural network learning.

Model collapse occurs when generative models degrade after repeatedly training on their own synthetic outputs. We study this effect in overparameterized linear regression in a setting where each iteration mixes fresh real labels with synthetic labels drawn from the model fitted in the previous iteration. We derive precise generalization error formulae for minimum-$\ell_2$-norm interpolation and ridge regression under this iterative scheme. Our analysis reveals intriguing properties of the optimal mixing weight that minimizes long-term prediction error and provably prevents model collapse. For instance, in the case of min-$\ell_2$-norm interpolation, we establish that the optimal real-data proportion converges to the reciprocal of the golden ratio for fairly general classes of covariate distributions. Previously, this property was known only for ordinary least squares, and additionally in low dimensions. For ridge regression, we further analyze two popular model classes -- the random-effects model and the spiked covariance model -- demonstrating how spectral geometry governs optimal weighting. In both cases, as well as for isotropic features, we uncover that the optimal mixing ratio should be at least one-half, reflecting the necessity of favoring real-data over synthetic. We study three additional settings: (i) where real data is fixed and fresh labels are not obtained at each iteration, (ii) where covariates vary across iterations but fresh real labels are available each time, and (iii) where covariates vary with time but only a fraction of them receive fresh real labels at each iteration. Across these diverse settings, we characterize when model collapse is inevitable and when synthetic data improves learning. We validate our theoretical results with extensive simulations.

We study the convergence rate of learning pairwise interactions in single-layer attention-style models, where tokens interact through a weight matrix and a nonlinear activation function. We prove that the minimax rate is $M^{-\frac{2\beta}{2\beta+1}}$, where $M$ is the sample size and $\beta$ is the H\"older smoothness of the activation function. Importantly, this rate is independent of the embedding dimension $d$, the number of tokens $N$, and the rank $r$ of the weight matrix, provided that $rd \le (M/\log M)^{\frac{1}{2\beta+1}}$. These results highlight a fundamental statistical efficiency of attention-style models, even when the weight matrix and activation are not separately identifiable, and provide a theoretical understanding of attention mechanisms and guidance on training.

The demand for efficient deployment of large language models (LLMs) has driven interest in quantization, which reduces inference cost, and parameter-efficient fine-tuning (PEFT), which lowers training overhead. This motivated the development of quantization-aware PEFT to produce accurate yet efficient quantized models. In this setting, reducing quantization error prior to fine-tuning is crucial for achieving high model accuracy. However, existing methods that rely on low-rank adaptation suffer from limited representational capacity. Recent Fourier-related transform (FT)-based adapters offer greater representational power than low-rank adapters, but their direct integration into quantized models often results in ineffective error reduction and increased computational overhead. To overcome these limitations, we propose QWHA, a method that integrates FT-based adapters into quantized models by employing the Walsh-Hadamard Transform (WHT) as the transform kernel, together with a novel adapter initialization scheme incorporating adaptive parameter selection and value refinement. We demonstrate that QWHA effectively mitigates quantization errors while facilitating fine-tuning, and that its design substantially reduces computational cost. Experimental results show that QWHA consistently outperforms baselines in low-bit quantization accuracy and achieves significant training speedups over existing FT-based adapters.

The statistical properties of deep neural networks (DNNs) at initialization play an important role to comprehend their trainability and the intrinsic architectural biases they possess before data exposure. Well established mean-field (MF) theories have uncovered that the distribution of parameters of randomly initialized networks strongly influences the behavior of the gradients, dictating whether they explode or vanish. Recent work has showed that untrained DNNs also manifest an initial-guessing bias (IGB), in which large regions of the input space are assigned to a single class. In this work, we provide a theoretical proof that links IGB to previous MF theories for a vast class of DNNs, showing that efficient learning is tightly connected to a network’s prejudice towards a specific class. This connection leads to a counterintuitive conclusion: the initialization that optimizes trainability is systematically biased rather than neutral.

We revisit Proximal Policy Optimization (PPO) from a function-space perspective. Our analysis decouples policy evaluation and improvement in a reproducing kernel Hilbert space (RKHS): (i) A kernelized temporal-difference (TD) critic performs efficient RKHS-gradient updates using only one-step state–action transition samples. (ii) a KL-regularized, natural-gradient policy step exponentiates the evaluated action-value, recovering a PPO/TRPO-style proximal update in continuous state-action spaces. We provide non-asymptotic, instance-adaptive guarantees whose rates depend on RKHS entropy, unifying tabular, linear, Sobolev, Gaussian, and Neural Tangent Kernel (NTK) regimes, and we derive a sampling rule for the proximal update that ensures the optimal $k^{-1/2}$ convergence rate for stochastic optimization. Empirically, the theory-aligned schedule improves stability and sample efficiency on common control tasks (e.g., CartPole, Acrobot, and HalfCheetah), while our TD-based critic attains favorable throughput versus a GAE baseline. Altogether, our results place PPO on a firmer theoretical footing beyond finite-dimensional assumptions and clarify when RKHS-proximal updates with kernel-TD critics yield global policy improvement with practical efficiency.

The LLM unlearning aims to eliminate the influence of undesirable data without affecting causally unrelated information. This process typically involves using a forget set to remove target information, alongside a retain set to maintain non-target capabilities. While recent localization-based methods demonstrate promise in identifying important nodes (neurons) to be unlearned, they fail to disentangle nodes responsible for forgetting undesirable knowledge or retaining essential skills, often treating them as a single entangled group. As a result, these methods apply uniform interventions, risking catastrophic over-forgetting or incomplete erasure of the target knowledge. To address this, we turn to circuit discovery, a mechanistic interpretability technique, and propose the Conflict-guided Localization for LLM Unlearning framEwork (CLUE). This framework identifies the forget and retain circuit composed of important nodes, and then the circuits are transformed into conjunctive normal forms (CNF). The assignment of each node in the CNF satisfiability solution reveals whether it should be forgotten or retained. We then provide targeted fine-tuning strategies for different categories of nodes. Extensive experiments demonstrate that, compared to existing localization methods, CLUE achieves superior forget efficacy and retain utility through precise neural localization.

As Vision Transformers (ViTs) become standard backbones across vision, a mechanistic account of their computational phenomenology is now essential. Despite architectural cues that hint at dynamical structure, there is no settled framework that interprets Transformer depth as a well-characterized flow. In this work, we introduce the $\textbf{Block-Recurrent Hypothesis (BRH)}$, arguing that trained ViTs admit a block-recurrent depth structure such that the computation of the original $L$ blocks can be accurately rewritten using only $k \ll L$ distinct blocks applied recurrently. Across diverse ViTs, between-layer representational similarity matrices suggest few contiguous phases. To determine whether this reflects reusable computation, we operationalize our hypothesis in the form of block recurrent surrogates of pretrained ViTs, which we call Recurrent Approximations to Phase-structured TransfORmers ($\texttt{Raptor}$). Using small-scale ViTs, we demonstrate that phase-structure metrics correlate with our ability to accurately fit $\texttt{Raptor}$ and identify the role of stochastic depth in promoting the recurrent block structure. We then provide an empirical existence proof for BRH in foundation models by showing that we can train a $\texttt{Raptor}$ model to recover $94$\% of DINOv2 ImageNet-1k linear probe accuracy in only 2 blocks. To provide a mechanistic account of these observations, we leverage our hypothesis to develop a program of $\textbf{Dynamical Interpretability}$. We find $\textit{\textbf{(i)}}$ directional convergence into class-dependent angular basins with self-correcting trajectories under small perturbations $\textit{\textbf{(ii)}}$ token-specific dynamics, where $\texttt{cls}$ executes sharp late reorientations while $\texttt{patch}$ tokens exhibit strong late-stage coherence reminiscent of a mean-field effect and converge rapidly toward their mean direction and $\textit{\textbf{(iii)}}$ a collapse of the update field to low rank in late depth, consistent with convergence to low-dimensional attractors. Altogether, we find that a compact recurrent program emerges along the depth of ViTs, pointing to a low-complexity normative solution that enables these models to be studied through principled dynamical systems analysis.

Prior work suggests that language models, while trained on next token prediction, show implicit planning behavior: they may select the next token in preparation to a predicted future token, such as a likely rhyming word, as supported by a prior qualitative study of Claude 3.5 Haiku using a cross-layer transcoder. We propose much simpler techniques for assessing implicit planning in language models. With case studies on rhyme poetry generation and question answering, we demonstrate that our methodology easily scales to many models. Across models, we find that the generated rhyme (e.g. "-ight") or answer to a question ("whale") can be manipulated by steering at the end of the preceding line with a vector, affecting the generation of intermediate tokens leading up to the rhyme or answer word. We show that implicit planning is a universal mechanism, present in smaller models than previously thought, starting from 1B parameters. Our methodology offers a widely applicable direct way to study implicit planning abilities of LLMs. More broadly, understanding planning abilities of language models can inform decisions in AI safety and control.

Attribution methods seek to explain language model predictions by quantifying the contribution of input tokens to generated outputs. However, most existing techniques are designed for encoder-based architectures and rely on linear approximations that fail to capture the causal and semantic complexities of autoregressive generation in decoder-only models. To address these limitations, we propose Hessian-Enhanced Token Attribution (HETA), a novel attribution framework tailored for decoder-only language models. HETA combines three complementary components: a semantic transition vector that captures token-to-token influence across layers, Hessian-based sensitivity scores that model second-order effects, and KL divergence to measure information loss when tokens are masked. This unified design produces context-aware, causally faithful, and semantically grounded attributions. Additionally, we introduce a curated benchmark dataset for systematically evaluating attribution quality in generative settings. Empirical evaluations across multiple models and datasets demonstrate that HETA consistently outperforms existing methods in attribution faithfulness and alignment with human annotations, establishing a new standard for interpretability in autoregressive language models.

A key component of in-context reasoning is the ability of language models (LMs) to bind entities for later retrieval. For example, an LM might represent Ann loves pie by binding Ann to pie, allowing it to later retrieve Ann when asked Who loves pie? Prior research on short lists of bound entities found strong evidence that LMs implement such retrieval via a positional mechanism, where Ann is retrieved based on its position in context. In this work, we find that this mechanism generalizes poorly to more complex settings; as the number of bound entities in context increases, the positional mechanism becomes noisy and unreliable in middle positions. To compensate for this, we find that LMs supplement the positional mechanism with a lexical mechanism (retrieving Ann using its bound counterpart pie) and a reflexive mechanism (retrieving Ann through a direct pointer). Through extensive experiments on nine models and ten binding tasks, we uncover a consistent pattern in how LMs mix these mechanisms to drive model behavior. We leverage these insights to develop a causal model combining all three mechanisms that estimates next token distributions with 95\% agreement. Finally, we show that our model generalizes to substantially longer inputs of open-ended text interleaved with entity groups, further demonstrating the robustness of our findings in more natural settings. Overall, our study establishes a more complete picture of how LMs bind and retrieve entities in-context.

Video generative models achieve high-quality synthesis from natural-language prompts by leveraging large-scale web data. However, this training paradigm inherently exposes them to unsafe biases and harmful concepts, introducing the risk of generating undesirable or illicit content. To mitigate unsafe generations, existing machine unlearning approaches either rely on filtering, and can therefore be bypassed, or they update model weights, but with costly fine-tuning or training-free closed-form edits. We propose the first training-free weight update framework for concept removal in video diffusion models. From five paired safe/unsafe prompts, our method estimates a refusal vector and integrates it into the model weights as a closed-form update. A contrastive low-rank factorization further disentangles the target concept from unrelated semantics, it ensures a selective concept suppression and it does not harm generation quality. Our approach reduces unsafe generations on the Open-Sora and ZeroScopeT2V models across the T2VSafetyBench and SafeSora benchmarks, with average reductions of 36.3% and 58.2% respectively, while preserving prompt alignment and video quality. This establishes an efficient and scalable solution for safe video generation without retraining nor any inference overhead.

Large Language Models (LLMs) have recently been successfully applied to regression tasks---such as time series forecasting and tabular prediction---by leveraging their in-context learning abilities. However, their autoregressive decoding process may be ill-suited to continuous-valued outputs, where obtaining predictive distributions over numerical targets requires repeated sampling, leading to high computational cost and inference time. In this work, we investigate whether distributional properties of LLM predictions can be recovered without explicit autoregressive generation. To this end, we study a set of regression probes trained to predict statistical functionals (e.g., mean, median, quantiles) of the LLM’s numerical output distribution directly from its internal representations. Our results suggest that LLM embeddings carry informative signals about summary statistics of their predictive distributions, including the numerical uncertainty. This investigation opens up new questions about how LLMs internally encode uncertainty in numerical tasks, and about the feasibility of lightweight alternatives to sampling-based approaches for uncertainty-aware numerical predictions.

Chain-of-Thought (CoT) reasoning often fails to faithfully reflect a language model's underlying decision process. We address this by introducing a *Markovian* language model framework with an autoencoder-style *reasoning bottleneck*: all information flowing from question to answer must pass through a bounded-length CoT, creating a bandwidth bottleneck analogous to the latent layer of an autoencoder. In practice, the KL penalty toward the pretrained distribution and the inductive biases of gradient descent discourage steganographic encoding, so the model learns to express its reasoning in natural-language steps from which the answer can be derived. We train this system with a GRPO-style policy gradient algorithm using parallel sampling, a frozen baseline CoT$'$, within-batch standardized advantages, and actor-reward (chain-rule) gradients. On QA tasks, Markovian training recovers most of the gains of a Non-Markovian GRPO variant while forcing the model to answer from the CoT alone (e.g., GSM8K: 19.6\% $\to$ 57.1\%; ARC-Challenge: 36.1\% $\to$ 79.9\%; on average within $\approx$3--4 pp of a Non-Markovian variant). Perturbation analyses across types and severities show that Markovian models incur systematically larger log-probability drops under CoT corruption than matched Non-Markovian baselines, indicating stronger causal reliance on the CoT. Cross-model evaluation confirms that learned CoTs generalize across architectures, suggesting they encode transferable reasoning steps rather than model-specific artifacts.

Do large language models (LLMs) represent concepts abstractly, i.e., independent of input format? We revisit Function Vectors (FVs), compact representations of in-context learning (ICL) tasks that causally drive task performance. Across multiple LLMs, we show that FVs are not fully invariant: FVs are nearly orthogonal when extracted from different input formats (e.g., open-ended vs. multiple-choice), even if both target the same concept. We identify Concept Vectors (CVs), which carry more stable concept representations. Like FVs, CVs are composed of attention head outputs; however, unlike FVs, the constituent heads are selected using Representational Similarity Analysis (RSA) based on whether they encode concepts consistently across input formats. While these heads emerge in similar layers to FV-related heads, the two sets are largely distinct, suggesting different underlying mechanisms. Steering experiments reveal that FVs excel in-distribution, when extraction and application formats match (e.g., both open-ended in English), while CVs generalize better out-of-distribution across both question types (open-ended vs. multiple-choice) and languages. Our results show that LLMs do contain abstract concept representations, but these differ from those that drive ICL performance.

LLMs require efficient knowledge editing (KE) to update factual information, yet existing methods exhibit significant performance decay in multi-hop factual recall. This failure is particularly acute when edits involve intermediate implicit subjects within reasoning chains. Through causal analysis, we reveal that this limitation stems from an oversight of how chained knowledge is dynamically represented and utilized at the neuron level. We discover that during multi-hop reasoning, implicit subjects function as query neurons, which sequentially activate corresponding value neurons across transformer layers to accumulate information toward the final answer—a dynamic prior KE work has overlooked. Guided by this insight, we propose ACE (Attribution-Controlled Knowledge Editing), a framework that leverages neuron-level attribution to identify and edit these critical query-value (Q-V) pathways. Ace provides a mechanistically grounded solution for multi-hop KE, empirically outperforming state-of-the-art methods by 9.44% on GPT-J and 37.46% on Qwen3-8B. Our analysis further reveals more fine-grained activation patterns in Qwen3 and demonstrates that the semantic interpretability of value neurons is orchestrated by query-driven accumulation. These findings establish a new pathway for advancing KE capabilities based on the principled understanding of internal reasoning mechanisms.

Sparse autoencoders (SAEs) have recently emerged as a powerful tool for interpreting the features learned by large language models (LLMs). By reconstructing features with sparsely activated networks, SAEs aim to recover complex superposed polysemantic features into interpretable monosemantic ones. Despite their wide applications, it remains unclear under what conditions SAEs can fully recover the ground truth monosemantic features from the superposed polysemantic ones. In this paper, we provide the first theoretical analysis with a closed-form solution for SAEs, revealing that they generally fail to fully recover the ground truth monosemantic features unless the ground truth features are extremely sparse. To improve the feature recovery of SAEs in general cases, we propose a reweighting strategy targeting at enhancing the reconstruction of the ground truth monosemantic features instead of the observed polysemantic ones. We further establish a theoretical weight selection principle for our proposed weighted SAE (WSAE). Experiments across multiple settings validate our theoretical findings and demonstrate that our WSAE significantly improves feature monosemanticity and interpretability.

We present a systematic study of medical-domain interpretability in Large Language Models (LLMs). We study how the LLMs both represent and process medical knowledge through four different interpretability techniques: (1) UMAP projections of intermediate activations, (2) gradient-based saliency with respect to the model weights, (3) layer lesioning/removal and (4) activation patching. We present knowledge maps of five LLMs which show, at a coarse-resolution, where knowledge about patient's ages, medical symptoms, diseases and drugs is stored in the models. In particular for Llama3.3-70B, we find that most medical knowledge is processed in the first half of the model's layers. In addition, we find several interesting phenomena: (i) age is often encoded in a non-linear and sometimes discontinuous manner at intermediate layers in the models, (ii) the disease progression representation is non-monotonic and circular at certain layers of the model, (iii) in Llama, drugs cluster better by medical specialty rather than mechanism of action, especially for Llama and (iv) Gemma-27B and MedGemma-27B have activations that collapse at intermediate layers but recover by the final layers. These results can guide future research on fine-tuning, un-learning or de-biasing LLMs for medical tasks by suggesting at which layers in the model these techniques should be applied. We attached our source code to the paper for reproducibility.

Understanding how populations of neurons represent information is a central challenge across machine learning and neuroscience. Recent work in both fields has begun to characterize the representational geometry and functionality underlying complex distributed activity. For example, artificial neural networks trained on data with more features than neurons compress data by representing features non-orthogonally in so-called superposition. However, the effect of time (or memory), an additional capacity-constraining pressure, on underlying representational geometry in recurrent models is not well understood. Here, we study how memory demands affect representational geometry in recurrent neural networks (RNNs), introducing the concept of temporal superposition. We develop a theoretical framework in RNNs with linear recurrence trained on a delayed serial recall task to better understand how properties of the data, task demands, and network dimensionality lead to different representational strategies, and show that these insights generalize to nonlinear RNNs. Through this, we identify an effectively linear, dense regime and a sparse regime where RNNs utilize an interference-free space, characterized by a phase transition in the angular distribution of features and decrease in spectral radius. Finally, we analyze the interaction of spatial and temporal superposition to observe how RNNs mediate different representational tradeoffs. Overall, our work offers a mechanistic, geometric explanation of representational strategies RNNs learn, how they depend on capacity and task demands, and why.

This work investigates how large language models (LLMs) internally represent emotion by analyzing the geometry of their hidden-state space. Using a synthetic dataset of emotionally rewritten sentences, we identify a low-dimensional emotional manifold via singular value decomposition and show that emotional representations are directionally encoded, distributed across layers, and aligned with interpretable dimensions. These structures are stable across depth and generalize to eight real-world emotion datasets spanning five languages. Cross-domain alignment yields low error and strong linear probe performance, indicating a universal emotional subspace. Within this space, internal emotion perception can be steered while preserving semantics using a learned intervention module, with especially strong control for basic emotions across languages. These findings reveal a consistent and manipulable affective geometry in LLMs and offer insight into how they internalize and process emotion.

We propose the Lattice Representation Hypothesis of large language models: a symbolic backbone that grounds conceptual hierarchies and logical operations in embedding geometry. Our framework unifies the Linear Representation Hypothesis with Formal Concept Analysis (FCA), showing that linear attribute directions with separating thresholds induce a concept lattice via half-space intersections. This geometry enables symbolic reasoning through geometric meet (intersection) and join (union) operations, and admits a canonical form when attribute directions are linearly independent. Experiments on WordNet sub-hierarchies provide empirical evidence that LLM embeddings encode concept lattices and their logical structure, revealing a principled bridge between continuous geometry and symbolic abstraction.

While reinforcement learning (RL) demonstrated remarkable success in enhancing the reasoning capabilities of language models, the training dynamics of RL in LLMs remain unclear. In this work, we provide an explanation of the RL training process through empirical analysis and rigorous theoretical modeling. First, through systematic reasoning-pattern-level and token-level analysis across the RL training process, we show that while different reasoning patterns exhibit relatively stable success rates during training, RL primarily optimizes a sparse subset of critical tokens, thereby reshaping reasoning pattern distributions to affect model performance. Building on these empirical insights, we develop a theoretical framework to understand the training dynamics of RL with two typical rewards: verifiable reward (RLVR) and model's internal feedback (RLIF). For RLVR, we analyze the training dynamics under two special cases: one where models readily converge to optimal reasoning strategies, and another where optimization becomes challenging, revealing that the base model's reasoning quality is crucial for determining convergence behavior. For RLIF, we examine how internal rewards initially improve model performance but can potentially lead to degradation with continued training. Extensive experiments validate our findings, advancing both theoretical understanding and practical applications of RL in language model enhancement.

The minimum-cost bipartite matching between two sets of points $R$ and $S$ in a metric space has a wide range of applications in machine learning, computer vision, and logistics. For instance, it can be used to estimate the $1$-Wasserstein distance between continuous probability distributions and for efficiently matching requests to servers while minimizing cost. However, the computational cost of determining the minimum-cost matching for general metrics spaces, poses a significant challenge, particularly in dynamic settings where points arrive over time and each update requires re-executing the algorithm. In this paper, given a fixed set $S$, we describe a deterministic algorithm that maintains, after $i$ additions to $R$, an $O(1/\delta^{0.631})$-approximate minimum-cost matching of cardinality $i$ between sets $R$ and $S$ in any metric space, with an amortized insertion time of $\widetilde{O}(n^{1+\delta})$ for adding points in $R$. To the best of our knowledge, this is the first algorithm for incremental minimum-cost matching that applies to arbitrary metric spaces. Interestingly, an important subroutine of our algorithm lends itself to efficient parallelization. We provide both a CPU implementation and a GPU implementation that leverages parallelism. Extensive experiments on both synthetic and real world datasets showcase that our algorithm either matches or outperforms all benchmarks in terms of speed while significantly improving upon the accuracy.

Heuristic design with large language models (LLMs) has emerged as a promising approach for tackling combinatorial optimization problems (COPs). However, existing approaches often rely on manually predefined evolutionary computation (EC) heuristic-optimizers and single-task training schemes, which may constrain the exploration of diverse heuristic algorithms and hinder the generalization of the resulting heuristics. To address these issues, we propose Meta-Optimization of Heuristics (MoH), a novel framework that operates at the optimizer level, discovering effective heuristic-optimizers through the principle of meta-learning. Specifically, MoH leverages LLMs to iteratively refine a meta-optimizer that autonomously constructs diverse heuristic-optimizers through (self-)invocation, thereby eliminating the reliance on a predefined EC heuristic-optimizer. These constructed heuristic-optimizers subsequently evolve heuristics for downstream tasks, enabling broader heuristic exploration. Moreover, MoH employs a multi-task training scheme to promote its generalization capability. Experiments on classic COPs demonstrate that MoH constructs an effective and interpretable meta-optimizer, achieving state-of-the-art performance across various downstream tasks, particularly in cross-size settings. Our code is available at: \url{https://github.com/yiding-s/MoH}.

Data pruning (DP), as an oft-stated strategy to alleviate heavy training burdens, reduces the volume of training samples according to a well-defined pruning method while striving for near-lossless performance. However, existing approaches, which commonly select highly informative samples, can lead to biased gradient estimation compared to full-dataset training. Furthermore, the analysis of this bias and its impact on final performance remains ambiguous. To address these challenges, we propose OrderDP, a plug-and-play framework that aims to obtain stable, unbiased, and near-lossless training acceleration with theoretical guarantees. Specifically, OrderDP first randomly selects a subset and then chooses the top-$q$ samples, where unbiasedness is established with respect to a surrogate loss. This ensures that OrderDP conducts unbiased training in terms of the surrogate objective. We further establish convergence and generalization analyses, elucidating how OrderDP affects optimal performance and enables well-controlled acceleration while ensuring guaranteed final performance. Empirically, we evaluate OrderDP against comprehensive baselines on CIFAR-10, CIFAR-100, and ImageNet-1K, demonstrating competitive accuracy, stable convergence, and exact control---all with a simpler design and faster runtime, while reducing training cost by over 40\%. Delivering both strong performance and computational efficiency, our method serves as a robust and easily adaptable tool for data-efficient learning.

Let $(X,d)$ be a metric space and let $\mu,\nu$ be discrete probability distributions supported on finite point sets $A,B \subseteq X$. For any $p \in [1,\infty]$, the {\it $W_p$-distance} between $\mu$ and $\nu$, $W_p(\mu, \nu)$, is defined as the $p$-th root of the minimum cost of transporting all the probability mass from $\mu$ to $\nu$, where moving a probability mass of $\delta$ from $a \in A$ to $b \in B$ incurs a cost of $\delta d(a,b)^p$. We give a (Las Vegas) randomized algorithm that computes a $(4+\varepsilon)$-approximate $W_p$ optimal-transport (OT) plan in $O(n^2 + (n^{3/2}\varepsilon^{-1}\log n\log\Delta)^{1+o(1)}\log U)$ time with probability at least $1-1/n$, for all $p \in [1,\infty]$, where $\varepsilon > 0$ is an arbitrarily small constant and $\Delta$ is the ratio between the largest and smallest interpoint distances in $A\cup B$. The previous best result achieved an $O(\log n)$-approximation in $O(pn^2)$ time, for constant values of $p$. Our algorithm significantly improves the approximation factor and, importantly, is the first quadratic-time method that extends to the $W_\infty$-distance. In contrast, additive approximation methods such as Sinkhorn are efficient only for constant $p$ and fail to handle $p=\infty$. \changed{Our algorithm also extends to a query model where, for any integer $k > 1$, we give an algorithm that preprocesses $X$ into clusters in $O(n^2+kn^{1+1/k}\log n\log\Delta)$ time, after which a $O(k)$-approximate $W_p$ distance between any two distributions $\mu$ and $\nu$ with $X$ as support can be computed in $(n^{1+1/k}\log n\log\Delta)^{1+o(1)}$ time with probability at most $1-1/n$.} Finally, for $p=\infty$, we show that obtaining a relative approximation factor better than $2$ in $O(n^2)$ time would resolve the long-standing open problem of computing a perfect matching in an arbitrary bipartite graph in quadratic time.

Large Language Models (LLMs) are increasingly employed in real-world applications, driving the need to evaluate the trustworthiness of their generated text. To this end, reliable uncertainty estimation is essential. Leading uncertainty estimation methods generate and analyze multiple output sequences, which is computationally expensive and impractical at scale. In this work, we inspect the theoretical foundations of these methods and explore new directions to enhance computational efficiency. Building on the framework of proper scoring rules, we find that the negative log-likelihood of the most likely output sequence constitutes a theoretically principled uncertainty measure. To approximate this alternative measure, we propose G-NLL, obtained using a single output sequence from greedy decoding. This approach streamlines uncertainty estimation while preserving theoretical rigor. Empirical results demonstrate that G-NLL achieves state-of-the-art performance across various scenarios. Our work lays the theoretical foundation for efficient and reliable uncertainty estimation in natural language generation, challenging the necessity of the prevalent methods that are more complex and resource-intensive.

We propose a general-purpose approach for improving the ability of large language models (LLMs) to intelligently and adaptively gather information from a user or other external source using the framework of sequential Bayesian experimental design (BED). This enables LLMs to act as effective multi-turn conversational agents and interactively interface with external environments. Our approach, which we call BED-LLM (Bayesian experimental design with large language models), is based on iteratively choosing questions or queries that maximize the expected information gain (EIG) with respect to a variable of interest given the responses gathered previously. We show how this EIG can be formulated (and then estimated) in a principled way using a probabilistic model derived from the LLM's predictive distributions and provide detailed insights into key decisions in its construction and updating procedure. We find that BED-LLM achieves substantial gains in performance across a wide range of tests based on the 20 Questions game and using the LLM to actively infer user preferences, compared to purely prompting-based design generation and other adaptive design strategies.

Noise Contrastive Estimation (NCE) has fueled major breakthroughs in representation learning and generative modeling. Yet a long-standing challenge remains: accurately estimating ratios between distributions that differ substantially, which significantly limits the applicability of NCE on modern high-dimensional and multimodal datasets. We revisit this problem from a less explored perspective: the magnitude of the noise distribution. Specifically, we show that with a virtually scaled (i.e., artificially increased) noise magnitude, the gradient of the NCE objective can closely align with that of Maximum Likelihood, enabling a trajectory-wise approximation from NCE to MLE, and faster convergence both theoretically and empirically. Building on this insight, we introduce "Noisier" NCE, a simple drop-in modification to vanilla NCE that incurs little to no extra computational cost, while effectively handling density-ratio estimation in challenging regimes where traditional MLE and NCE struggle. Beyond improving classical density-ratio learning, "Noisier" NCE proves broadly applicable: it achieves strong results across image modeling, anomaly detection, and offline black-box optimization. On CIFAR-10 and ImageNet64×64 datasets, it yields 10-step and even 1-step samplers that match or surpass state-of-the-art methods, while cutting training iterations by up to half.

Computing next-token likelihood ratios between two language models (LMs) is a standard task in training paradigms such as knowledge distillation. Since this requires both models to share the same probability space, it becomes challenging when the teacher and student LMs use different tokenizers, for instance, when edge-device deployment necessitates a smaller vocabulary size to lower memory overhead. This work addresses this vocabulary misalignment problem by uncovering an implicit recursive structure in the commonly deployed Byte-Pair Encoding (BPE) algorithm and utilizing it to create a probabilistic framework for \textit{cross-tokenizer likelihood scoring}. Our method enables sequence likelihood evaluation for vocabularies different from the teacher model native tokenizer, addressing two specific scenarios: when the student vocabulary is a subset of the teacher vocabulary, and the general case where it is arbitrary. In the subset regime, our framework computes exact likelihoods and provides next-token probabilities for sequential sampling with only $\mathcal{O}(1)$ model evaluations per token. When used for distillation, this yields up to a 12% reduction in memory footprint for the Qwen2.5-1.5B model while also improving baseline performance up to 4\% on the evaluated tasks. For the general case, we introduce a rigorous lossless procedure that leverages BPE recursive structure, complemented by a fast approximation that keeps large-vocabulary settings practical. Applied to GSM8K mathematical reasoning distillation, our method improves accuracy by over 2% the current state of the art. Code: https://github.com/truongbuu/cross-tokenizer-scoring

Sliced Mutual Information (SMI) is widely used as a scalable alternative to mutual information for measuring non-linear statistical dependence. Despite its advantages, such as faster convergence, robustness to high dimensionality, and nullification only under statistical independence, we demonstrate that SMI is highly susceptible to data manipulation and exhibits counterintuitive behavior. Through extensive benchmarking and theoretical analysis, we show that SMI saturates easily, fails to detect increases in statistical dependence (even under linear transformations designed to enhance the extraction of information), prioritizes redundancy over informative content, and in some cases, performs worse than simpler dependence measures like the correlation coefficient.

Deploying reinforcement learning (RL) in robotics, industry, and health care is blocked by two obstacles: the difficulty of specifying accurate rewards and the risk of unsafe, data-hungry exploration. We address this by proposing a two-stage framework that first learns a safe initial policy from a reward-free dataset of expert demonstrations, then fine-tunes it online using preference-based human feedback. We provide the first principled analysis of this offline-to-online approach and introduce BRIDGE, a unified algorithm that integrates both signals via an uncertainty-weighted objective. We derive regret bounds that shrink with the number of offline demonstrations, explicitly connecting the quantity of offline data to online sample efficiency. We validate BRIDGE in discrete and continuous control MuJoCo environments, showing it achieves lower regret than both standalone behavioral cloning and online preference-based RL. Our work establishes a theoretical foundation for designing more sample-efficient interactive agents.

Foundation MLIPs demonstrate broad applicability across diverse material systems and have emerged as a powerful and transformative paradigm in chemical and computational materials science. Equivariant MLIPs achieve state-of-the-art accuracy in a wide range of benchmarks by incorporating equivariant inductive bias. However, the reliance on tensor products and high-degree representations makes them computationally costly. This raises a fundamental question: as quantum mechanical-based datasets continue to expand, can we develop a more compact model to thoroughly exploit high-dimensional atomic interactions? In this work, we present MatRIS (\textbf{Mat}erials \textbf{R}epresentation and \textbf{I}nteraction \textbf{S}imulation), an invariant MLIP that introduces attention-based modeling of three-body interactions. MatRIS leverages a novel separable attention mechanism with linear complexity $O(N)$, enabling both scalability and expressiveness. MatRIS delivers accuracy comparable to that of leading equivariant models on a wide range of popular benchmarks (Matbench-Discovery, MatPES, MDR phonon, Molecular dataset, etc). Taking Matbench-Discovery as an example, MatRIS achieves an F1 score of up to 0.847 and attains comparable accuracy at a lower training cost. The work indicates that our carefully designed invariant models can match or exceed the accuracy of equivariant models at a fraction of the cost, shedding light on the development of accurate and efficient MLIPs.

Bayesian posterior sampling techniques have demonstrated superior empirical performance in many exploration-exploitation settings. However, their theoretical analysis remains a challenge, especially in complex settings like reinforcement learning. In this paper, we introduce Q-Learning with Posterior Sampling (PSQL), a simple Q-learning-based algorithm that uses Gaussian posteriors on Q-values for exploration, akin to the popular Thompson Sampling algorithm in the multi-armed bandit setting. We show that in the tabular episodic MDP setting, PSQL achieves a regret bound of $\tilde O(H^2\sqrt{SAT})$, closely matching the known lower bound of $\Omega(H\sqrt{SAT})$. Here, S, A denote the number of states and actions in the underlying Markov Decision Process (MDP), and $T=KH$ with $K$ being the number of episodes and $H$ being the planning horizon. Our work provides several new technical insights into the core challenges in combining posterior sampling with dynamic programming and TD-learning-based RL algorithms, along with novel ideas for resolving those difficulties. We hope this will form a starting point for analyzing this efficient and important algorithmic technique in even more complex RL settings.

Policy gradient methods are known to be highly sensitive to the choice of policy parameterization. In particular, the widely used softmax parameterization can induce ill-conditioned optimization landscapes and lead to exponentially slow convergence. Although this can be mitigated by preconditioning, this solution is often computationally expensive. Instead, we propose replacing the softmax with an alternative family of policy parameterizations based on the generalized $f$-$\textit{softargmax}$. We further advocate coupling this parameterization with a regularizer induced by the same $f$-divergence, which improves the optimization landscape and ensures that the resulting regularized objective satisfies a Polyak--Łojasiewicz inequality. Leveraging this structure, we establish the $\textit{first explicit non-asymptotic last-iterate convergence guarantees}$ for stochastic policy gradient methods for finite MDPs $\textit{without any form of preconditioning}$. We also derive sample-complexity bounds for the unregularized problem and show that $f$-PG, with Tsallis divergences achieves $\textit{polynomial sample complexity}$ in contrast to the exponential complexity incurred by the standard softmax parameterization.

We prove the stability and global convergence of a coupled actor-critic gradient flow for infinite-horizon and entropy-regularised Markov decision processes (MDPs) in continuous state and action space with linear function approximation under Q-function realisability. We consider a version of the actor critic gradient flow where the critic is updated using temporal difference (TD) learning while the policy is updated using a policy mirror descent method on a separate timescale. For general action spaces, the relative entropy regularizer is unbounded and thus it is not clear a priori that the actor-critc flow does not suffer from finite-time blow-up. Therefore we first demonstrate stability which in turn enables us obtain a convergence rate of the actor critic flow to the optimal regularised value function. The arguments presented show that timescale separation is crucial for stability and convergence in this setting.

Training web agents to navigate complex, real-world websites requires them to master subtasks—short-horizon interactions on multiple UI components (e.g., choosing the correct date in a date picker, or scrolling in a container to extract information). We introduce WARC-Bench (Web Archive Benchmark), a novel web navigation benchmark featuring 438 tasks designed to evaluate multimodal AI agents on subtasks. WARC-Bench enables sandboxed interactions with dynamic and realistic webpages using Web ARChive files. We show that WARC-Bench is challenging for leading computer-use models, with the highest observed success rate being 64.8%. To improve open source models on subtask, we explore two common training techniques: supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR). Experiments show that SFT models obtain a 48.8% success rate on the benchmark. Training with RLVR over SFT checkpoints, even in data-scarce settings, improves the score to 52.8% on WARC-Bench, outperforming many frontier models. Our analysis concludes that mastering these subtasks is essential for robust web planning and navigation, and is a capability not extensively evaluated by existing benchmarks. More details about WARC-Bench can be found at https://sanjari-orb.github.io/warc-bench/.

Off-policy evaluation (OPE) is a fundamental task in reinforcement learning (RL). In the classic setting of \emph{linear OPE}, finite-sample guarantees often take the form $$ \textrm{Prediction error} \le \textrm{poly}(C^\pi, d, 1/n, log(1/\delta)), $$ where $d$ is the dimension of the features, and $C^\pi$ is a **_feature coverage parameter_** that characterizes the degree to which the visited features lie in the span of the data distribution. While such guarantees are well-understood for several popular algorithms under stronger assumptions (e.g. Bellman completeness), the understanding is lacking and fragmented in the minimal setting where the target value function is linearly realizable in the features. Despite recent interest in tight characterizations of the statistical rate in this setting, the right notion of coverage remains unclear, and candidate definitions from prior analyses have undesirable properties and are starkly disconnected from more standard definitions in the literature. We provide a novel finite-sample analysis of a canonical algorithm for this setting, LSTDQ. Inspired by an instrumental-variable view, we develop error bounds that depend on a novel coverage parameter, the **feature-dynamics coverage**, which can be interpreted as linear coverage in an induced dynamical system for feature evolution. With further assumptions---such as Bellman-completeness---our definition successfully recovers the coverage parameters specialized to those settings, finally yielding a unified understanding for coverage in linear OPE.

Reinforcement learning for safety-critical tasks requires policies that are both high-performing and safe throughout the learning process. While model-predictive shielding is a promising approach, existing methods are often computationally intractable for the high-dimensional, nonlinear systems where deep RL excels, as they typically rely on a patchwork of local models. We introduce RAMPS, a scalable shielding framework that overcomes this limitation by leveraging a learned, linear representation of the environment's dynamics. This model can range from a linear regression in the original state space to a more complex operator learned in a high-dimensional feature space. The key is that this linear structure enables a robust, look-ahead safety technique based on a multi-step Control Barrier Function (CBF). By moving beyond myopic one-step formulations, RAMPS accounts for model error and control delays to provide reliable, real-time interventions. The resulting framework is minimally invasive, computationally efficient, and built upon robust control-theoretic foundations. Our experiments demonstrate that RAMPS significantly reduces safety violations compared to existing safe RL methods while maintaining high task performance in complex control environments.

A core challenge of Monte Carlo Tree Search (MCTS) is its sample efficiency, which can be addressed by building and using state and/or state-action pair abstractions in parallel to the tree search, such that information can be shared among nodes of the same layer. On the Go Abstractions in Upper Confidence bounds applied to Trees (OGA-UCT) is the state-of-the-art MCTS abstraction algorithm for deterministic environments that builds its abstraction using the Abstractions of State-Action Pairs (ASAP) framework, which aims to detect states and state-action pairs with the same value under optimal play by analysing the search graph. ASAP, however, requires two state-action pairs to have the same immediate reward, which is a rigid condition that limits the number of abstractions that can be found and thereby the sample efficiency. In this paper, we break with the paradigm of grouping value-equivalent states or state-action pairs and instead group states and state-action pairs with possibly different values as long as the difference between their values can be inferred. We call this abstraction framework Known Value Difference Abstractions (KVDA), which infers the value differences by analysis of the immediate rewards and modifies OGA-UCT to use this framework instead. The modification is called KVDA-UCT, which detects significantly more abstractions than OGA-UCT, introduces no additional parameter, and outperforms OGA-UCT on a variety of deterministic environments and parameter settings.

Nearest Neighbor Search (NNS) is a fundamental problem in data structures with wide-ranging applications, such as web search, recommendation systems, and, more recently, retrieval-augmented generations (RAG). In such recent applications, in addition to the relevance (similarity) of the returned neighbors, diversity among the neighbors is a central requirement. In this paper, we develop principled welfare-based formulations in NNS for realizing diversity across attributes. Our formulations are based on welfare functions---from mathematical economics---that satisfy central diversity (fairness) and relevance (economic efficiency) axioms. With a particular focus on Nash social welfare, we note that our welfare-based formulations provide objective functions that adaptively balance relevance and diversity in a query-dependent manner. Notably, such a balance was not present in the prior constraint-based approach, which forced a fixed level of diversity and optimized for relevance. In addition, our formulation provides a parametric way to control the trade-off between relevance and diversity, providing practitioners with flexibility to tailor search results to task-specific requirements. We develop efficient nearest neighbor algorithms with provable guarantees for the welfare-based objectives. Notably, our algorithm can be applied on top of any standard ANN method (i.e., use standard ANN method as a subroutine) to efficiently find neighbors that approximately maximize our welfare-based objectives. Experimental results demonstrate that our approach is practical and substantially improves diversity while maintaining high relevance of the retrieved neighbors.

Spiking neural networks (SNNs) are a promising paradigm for energy-efficient computation, yet their theoretical foundations—especially regarding stability and robustness—remain limited compared to artificial neural networks. In this work, we study discrete-time leaky integrate-and-fire (LIF) SNNs through the lens of Boolean function analysis. We focus on noise sensitivity and stability in classification tasks, quantifying how input perturbations affect outputs. Our main result shows that wide LIF-SNN classifiers are stable on average, a property explained by the concentration of their Fourier spectrum on low-frequency components. Motivated by this, we introduce the notion of spectral simplicity, which formalizes simplicity in terms of Fourier spectrum concentration and connects our analysis to the simplicity bias observed in deep networks. Within this framework, we show that random LIF-SNNs are biased toward simple functions. Experiments on trained networks confirm that these stability properties persist in practice. Together, these results provide new insights into the stability and robustness properties of SNNs.

Quantizing the weights of large language models (LLMs) from 16-bit to lower bitwidth is the de facto approach to deploy massive transformers onto more affordable accelerators. While GPTQ emerged as one of the standard methods for one-shot post-training quantization at LLM scale, its inner workings are described as a sequence of algebraic updates that obscure geometric meaning or worst-case guarantees. In this work, we show that, when executed back-to-front (from the last to first dimension) for a linear layer, GPTQ is mathematically identical to Babai's nearest plane algorithm for the classical closest vector problem (CVP) on a lattice defined by the Hessian matrix of the layer's inputs. This equivalence is based on a sophisticated mathematical argument, and has two analytical consequences: first, the GPTQ error propagation step gains an intuitive geometric interpretation; second, GPTQ inherits the error upper bound of Babai's algorithm under the assumption that no weights are clipped. Leveraging this bound, we design post-training quantization methods that avoid clipping, and outperform the original GPTQ. In addition, we provide efficient GPU inference kernels for the resulting representation. Taken together, these results place GPTQ on a firm theoretical footing and open the door to importing decades of progress in lattice algorithms towards the design of future quantization algorithms for billion-parameter models. Source code is available at https://github.com/IST-DASLab/GPTQ-Babai.

Recent empirical studies have explored the idea of continuing to train a model at test-time for a given task, known as test-time training (TTT), and have found it to yield significant performance improvements. However, there is limited understanding of why and when TTT is effective. Earlier explanations mostly focused on the observation that TTT may help when applied to out-of-distribution adaptation or used with privileged data. However, the growing scale of foundation models with most test data being in-distribution questions these explanations. We instead posit that foundation models remain globally underparameterized, with TTT providing a mechanism for specialization after generalization—focusing capacity on concepts relevant to the test task. Specifically, under the linear representation hypothesis, we propose a model in which TTT achieves a substantially smaller in-distribution test error than global training. We empirically validate our model's key assumptions by training a sparse autoencoder on ImageNet, showing that semantically related data points are explained by only a few shared concepts. Finally, we perform scaling studies across image and language tasks that confirm the practical implications of our model, identifying the regimes where specialization is most effective. Beyond TTT, our results provide additional strong evidence in support of the linear representation hypothesis.

Return-conditioned supervised learning (RCSL) algorithms have demonstrated strong generative capabilities in offline reinforcement learning (RL) by learning action distributions based on both the state and the return. However, many existing approaches treat RL as a conditional sequence modeling task, which can lead to an overreliance on suboptimal past experiences, impairing decision-making and reducing the effectiveness of trajectory synthesis. To address these limitations, we propose a novel approach, the Q-Augmented Dual-Feature Fusion Decision Transformer (QDFFDT), which adaptively combines both global sequence features and local immediate features through a learnable fusion mechanism. This model improves generalization across different tasks without the need for extensive hyperparameter tuning. Experimental results on the D4RL benchmark show that QDFFDT outperforms current methods, establishing new state-of-the-art performance and demonstrating the power of adaptive feature fusion.

Reinforcement learning in combinatorial action spaces requires searching over exponentially many joint actions to simultaneously select multiple sub-actions that form coherent combinations. Existing approaches either simplify policy learning by assuming independence across sub-actions, which often yields incoherent or invalid actions when coordination is required, or attempt to learn action structure and control jointly, which is slow and unstable. We introduce Structured Policy Initialization (SPIN), a two-stage framework that first pre-trains an Action Structure Model (ASM) to capture the manifold of valid actions, then freezes this representation and trains lightweight policy heads for control. On challenging DM Control benchmarks, SPIN improves average return by up to $39\%$ over the state of the art while reducing time to convergence by up to $12.8\times$.

Offline reinforcement learning (RL) enables learning effective policies from fixed datasets without any environment interaction. Existing methods typically employ policy constraints to mitigate the distribution shift encountered during offline RL training. However, because the scale of the constraints varies across tasks and datasets of differing quality, existing methods must meticulously tune hyperparameters to match each dataset, which is time-consuming and often impractical. To bridge this gap, we propose Adaptive Scaling of Policy Constraints (ASPC), a second-order differentiable framework that automatically adjusts the scale of policy constraints during training. We theoretically analyze its performance improvement guarantee. In experiments on 39 datasets across four D4RL domains, ASPC using a single hyperparameter configuration outperforms other adaptive constraint methods and state-of-the-art offline RL algorithms that require per-dataset tuning, achieving an average 35\% improvement in normalized performance over the baseline. Moreover, ASPC consistently yields additional gains when integrated with a variety of existing offline RL algorithms, demonstrating its broad generality.

Incorporating demonstration data into reinforcement learning (RL) can greatly accelerate learning, but existing approaches often assume demonstrations are optimal and fully aligned with the target task. In practice, demonstrations are frequently sparse, suboptimal, or misaligned, which can degrade performance when these demonstrations are integrated into RL. We propose Adaptive Policy Composition (APC), a hierarchical model that adaptively composes multiple data-driven Normalizing Flow (NF) priors. Instead of enforcing strict adherence to the priors, APC estimates each prior's applicability to the target task while leveraging them for exploration. Moreover, APC either refines useful priors, or sidesteps misaligned ones when necessary to optimize downstream reward. Across diverse benchmarks, APC accelerates learning when demonstrations are aligned, remains robust under severe misalignment, and leverages suboptimal demonstrations to bootstrap exploration while avoiding performance ceilings caused by overly strict adherence to suboptimal demonstrations.

Offline reinforcement learning (RL) seeks to learn an optimal policy from a fixed dataset, but distributional shift between the dataset and the learned policy often leads to suboptimal real-world performance. Existing methods typically use behavior policy regularization to constrain the learned policy, but these conservative approaches can limit performance and generalization, especially when the behavior policy is suboptimal. We propose a Peer-Influenced Constraint (PIC) framework with a ``peer review" mechanism. Specifically, we construct a set of similar states and use the corresponding actions as candidates, from which we select the optimal action to constrain the policy. This method helps the policy escape local optima while approximately ensuring the staying within the in-distribution space, boosting both performance and generalization. We also introduce an improved version, Ensemble Peer-Influenced Constraint (EPIC), which combines ensemble methods to achieve strong performance while maintaining high efficiency. Additionally, we uncover the Coupling Effect between PIC and uncertainty estimation, providing valuable insights for offline RL. We evaluate our methods on classic continuous control tasks from the D4RL benchmark, with both PIC and EPIC achieving competitive performance compared to state-of-the-art approaches.

Bootstrapping bias problem is a long-standing challenge in temporal-difference (TD) methods in off-policy reinforcement learning (RL). Multi-step return backups can alleviate this issue but require delicate importance sampling to correct their off-policy bias. Recent work has proposed to use chunked critics, which estimate the value of short action sequences ("chunks") rather than individual actions, enabling unbiased multi-step backup. However, extracting policies from chunked critics is challenging: policies must output the entire action chunk open-loop, which can be sub-optimal in environments that require policy reactivity and also challenging to model especially when the chunk length grows. Our key insight is to decouple the chunk length of the critic from that of the policy, allowing the policy to operate over shorter action chunks. We propose a novel algorithm that achieves this by optimizing the policy against a distilled critic for partial action chunks, constructed by optimistically backing up from the original chunked critic to approximate the maximum value achievable when a partial action chunk is extended to a complete one. This design retains the benefits of multi-step value propagation while sidestepping both the open-loop sub-optimality and the difficulty of learning policies over long action chunks. We evaluate our method on challenging, long-horizon offline goal-conditioned benchmarks and show that it reliably outperforms prior methods.

Recent advancements in language models have demonstrated remarkable in-context learning abilities, prompting the exploration of in-context reinforcement learning (ICRL) to extend the promise to decision domains. Due to involving more complex dynamics and temporal correlations, existing ICRL approaches may face challenges in learning from suboptimal trajectories and achieving precise in-context inference. In the paper, we propose Scalable In-Context Q-Learning (S-ICQL), an innovative framework that harnesses dynamic programming and world modeling to steer ICRL toward efficient reward maximization and task generalization, while retaining the scalability and stability of supervised pretraining. We design a prompt-based multi-head transformer architecture that simultaneously predicts optimal policies and in-context value functions using separate heads. We pretrain a generalized world model to capture task-relevant information, enabling the construction of a compact prompt that facilitates fast and precise in-context inference. During training, we perform iterative policy improvement by fitting a state value function to an upper-expectile of the Q-function, and distill the in-context value functions into policy extraction using advantage-weighted regression. Extensive experiments across a range of discrete and continuous environments show consistent performance gains over various types of baselines, especially when learning from suboptimal data. Our code is available at https://github.com/NJU-RL/SICQL.

Model-based methods for offline Reinforcement Learning transfer extensive policy exploration and evaluation to data-driven dynamics models, effectively saving real-world samples in the offline setting. We expect the dynamics model to allow the policy to roll out full-horizon episodes, which is crucial for ensuring sufficient exploration and reliable evaluation. However, many previous dynamics models exhibit limited capability in long-horizon prediction. This work follows the paradigm of the Any-step Dynamics Model (ADM) that improves future predictions by reducing bootstrapping prediction to direct prediction. We structurally decouple each recurrent forward of the RNN cell from the backtracked state and propose the second version of ADM (ADM-v2), making the direct prediction more flexible. ADM-v2 not only enhances the accuracy of direct predictions for making full-horizon roll-outs but also supports parallel estimation of the any-step prediction uncertainty to improve efficiency. The results on DOPE validate the reliability of ADM-v2 for policy evaluation. Moreover, via full-horizon roll-out, ADM-v2 for policy optimization enables substantial advancements, whereas other dynamics models degrade due to long-horizon error accumulation. We are the first to achieve SOTA under the full-horizon roll-out setting on both D4RL and NeoRL. The code is available at https://github.com/LAMDA-RL/adm2.

We study the problem of offline preference-based reinforcement learning (PbRL), where the agent learns from pre-collected preference data by comparing trajectory pairs. While prior work has established theoretical foundations for offline PbRL, existing algorithms face significant practical limitations: some rely on computationally intractable optimization procedures, while others suffer from unstable training and high performance variance. To address these challenges, we propose Preference-based Value Optimization (PVO), a simple and practical algorithm that achieves both strong empirical performance and theoretical guarantees. PVO directly optimizes the value function consistent with preference feedback by minimizing a novel \emph{value alignment loss}. We prove that PVO attains a rate-optimal sample complexity of $\mathcal{O}(\varepsilon^{-2})$, and further show that the value alignment loss is applicable not only to value-based methods but also to actor–critic algorithms. Empirically, PVO achieves robust and stable performance across diverse continuous control benchmarks. It consistently outperforms strong baselines, including methods without theoretical guarantees, while requiring no additional hyperparameters for preference learning. Moreover, our ablation study demonstrates that substituting the standard TD loss with the value alignment loss substantially improves learning from preference data, confirming its effectiveness for PbRL.

Generative models have gained significant traction in offline reinforcement learning (RL) due to their ability to model complex trajectory distributions. However, existing generation-based approaches still struggle with long-horizon tasks characterized by sparse rewards. Some hierarchical generation methods have been developed to mitigate this issue by decomposing the original problem into shorter-horizon subproblems using one policy and generating detailed actions with another. While effective, these methods often overlook the multi-scale temporal structure inherent in trajectories, resulting in suboptimal performance. To overcome these limitations, we propose MAGE, a Multi-scale Autoregressive GEneration-based offline RL method. MAGE incorporates a condition-guided multi-scale autoencoder to learn hierarchical trajectory representations, along with a multi-scale transformer that autoregressively generates trajectory representations from coarse to fine temporal scales. MAGE effectively captures temporal dependencies of trajectories at multiple resolutions. Additionally, a condition-guided decoder is employed to exert precise control over short-term behaviors. Extensive experiments on five offline RL benchmarks against fifteen baseline algorithms show that MAGE successfully integrates multi-scale trajectory modeling with conditional guidance, generating coherent and controllable trajectories in long-horizon sparse-reward settings. The source code is available at https://github.com/xmu-rl-3dv/MAGE.

Reinforcement learning from human feedback (RLHF) with proximal policy optimization (PPO) is widely used but often yields less diverse outputs than supervised fine-tuning, suggesting an effect in which the policy’s support contracts during on-policy optimization. We formalize this “policy contraction” with the Support Retention Ratio (SRR)—the share of SFT completions that retain non-negligible probability under the RL policy—and additionally track token-entropy, Kullback–Leibler (KL) divergence to the reference, and repetition. We propose Contraction-Aware PPO (CaPPO), a minimum-norm multi-gradient update that co-optimizes reward, entropy, and KL, paired with a controller that steers exploration toward a target token entropy. On HH-RLHF, Summarize-from-Feedback, and UltraFeedback with Qwen2-7B, Qwen2.5-14B, Mistral-7B-Instruct, and Llama-3-8B-Instruct, CaPPO increases win rate by 2 to 4 points over PPO and improves diversity, gaining 0.2 to 0.3 higher SRR. The gains persist under decoding sweeps and are robust to reward scaling and critic variance. Treating reward, diversity, and stability as first-class objectives, CaPPO mitigates contraction without sacrificing alignment performance.

The use of target networks is a popular approach for estimating value functions in deep Reinforcement Learning (RL). While effective, the target network remains a compromise solution that preserves stability at the cost of slowly moving targets, thus delaying learning. Conversely, using the online network as a bootstrapped target is intuitively appealing, albeit well-known to lead to unstable learning. In this work, we aim to obtain the best out of both worlds by introducing a novel update rule that computes the target using the MINimum estimate between the Target and Online network, giving rise to our method, MINTO. Through this simple, yet effective modification, we show that MINTO enables faster and stable value function learning, by mitigating the potential overestimation bias of using the online network for bootstrapping. Notably, MINTO can be seamlessly integrated into a wide range of value-based and actor-critic algorithms with a negligible cost. We evaluate MINTO extensively across diverse benchmarks, spanning online and offline RL, as well as discrete and continuous action spaces. Across all benchmarks, MINTO consistently improves performance, demonstrating its broad applicability and effectiveness.

Leveraging offline data is a promising way to improve the sample efficiency of online reinforcement learning (RL). This paper expands the pool of usable data for offline-to-online RL by leveraging abundant non-curated data that is reward-free, of mixed quality, and collected across multiple embodiments. Although learning a world model appears promising for utilizing such data, we find that naive fine-tuning fails to accelerate RL training on many tasks. Through careful investigation, we attribute this failure to the distributional shift between offline and online data during fine-tuning. To address this issue and effectively use the offline data, we propose two techniques: i) experience rehearsal and ii) execution guidance. With these modifications, the non-curated offline data substantially improves RL's sample efficiency. Under limited sample budgets, our method achieves nearly twice the aggregate score of learning-from-scratch baselines across 72 visuomotor tasks spanning 6 embodiments. On challenging tasks such as locomotion and robotic manipulation, it outperforms prior methods that utilize offline data by a decent margin.

Large Language Model (LLM) agents show great promise for complex multi-turn tool-use tasks, but their development is often hampered by the extreme scarcity of high-quality training data. Supervised fine-tuning (SFT) on synthetic data leads to overfitting, whereas standard reinforcement learning (RL) struggles with a critical cold-start problem and training instability. To address these challenges, we introduce $\textbf{Environment Tuning}$, a novel training paradigm that enables agents to learn complex behaviors directly from problem instances without relying on pre-collected expert trajectories. $\textbf{Environment Tuning}$ orchestrates this learning process through a structured curriculum, actionable environment augmentation that provides corrective feedback, and fine-grained progress rewards to ensure stable and efficient exploration. Using only 400 problem instances from Berkeley Function-Calling Leaderboard (BFCL) benchmark, our method not only achieves competitive in-distribution performance against strong baselines but also demonstrates superior out-of-distribution generalization, overcoming the performance collapse common to SFT-based approaches. Our work presents a paradigm shift from supervised fine-tuning on static trajectories to dynamic, environment-based exploration, paving the way for training more robust and data-efficient agents.

Inspired by the human learning and memory system, particularly the interplay between the hippocampus and cerebral cortex, this study proposes a dual-learner framework comprising a fast learner and a meta learner to address continual Reinforcement Learning~(RL) problems. These two learners are coupled to perform distinct yet complementary roles: the fast learner focuses on knowledge transfer, while the meta learner ensures knowledge integration. In contrast to traditional multi-task RL approaches that share knowledge through average return maximization, our meta learner incrementally integrates new experiences by explicitly minimizing catastrophic forgetting, thereby supporting efficient cumulative knowledge transfer for the fast learner. To facilitate rapid adaptation in new environments, we introduce an adaptive meta warm-up mechanism that selectively harnesses past knowledge. We conduct experiments in various pixel-based and continuous control benchmarks, revealing the superior performance of continual learning for our proposed dual-learner approach relative to baseline methods.

We propose FlowRL: matching the full reward distribution via flow balancing instead of solely maximizing rewards in large language model (LLM) reinforcement learning (RL). Recent advanced reasoning models adopt reward-maximizing methods (e.g., PPO and GRPO), which tend to over-optimize dominant reward signals while neglecting less frequent but valid reasoning paths, thus reducing diversity. In contrast, we transform scalar rewards into a normalized target distribution using a learnable partition function, and then minimize the reverse KL divergence between the policy and the target distribution. We implement this idea as a flow-balanced optimization method that promotes diverse exploration and generalizable reasoning trajectories. We conduct experiments on both math and code reasoning tasks: FlowRL achieves a significant average improvement of $10.0\%$ over GRPO and $5.1\%$ over PPO on math benchmarks, and performs consistently better on code reasoning tasks. These results highlight reward distribution-matching as a key step toward efficient exploration and diverse reasoning in LLM reinforcement learning.

The challenge of coalition formation games lies in efficiently navigating the exponentially large space of possible coalitions to identify the optimal partition. While existing approaches to solve coalition formation games either provide optimal solutions with limited scalability or approximate solutions without quality guarantees, we propose a novel scalable and sample-efficient approximation method based on deep reinforcement learning. Specifically, we model the coalition formation problem as a finite Markov decision process and use deep neural network to approximate optimal coalition structures within the full and abstracted coalition space. Moreover, our method is applicable to bi-level optimization problems in which coalition values are determined by the policies of individual agents at a lower decision-making level. This way, our approach facilitates dynamic, adaptive adjustments to coalition value assessments as they evolve over time. Empirical results demonstrate our algorithm's effectiveness in approximating optimal coalition structures in both normal-form and sequential mixed-motive games.

Large-scale pre-training has fundamentally changed how machine learning research is done today: large foundation models are trained once, and then can be used by anyone in the community (including those without data or compute resources to train a model from scratch) to adapt and fine-tune to specific tasks. Applying this same framework to reinforcement learning (RL) is appealing because it offers compelling avenues for addressing core challenges in RL, including sample efficiency and robustness. However, there remains a fundamental challenge to pre-train large models in the context of RL: actions have long-term dependencies, so training a foundation model that reasons across *time* is important. Recent advances in generative AI have provided new tools for modeling highly complex distributions. In this paper, we build a probabilistic model to predict which states an agent will visit in the temporally distant future (i.e., an occupancy measure) using flow matching. As large datasets are often constructed by many distinct users performing distinct tasks, we include in our model a latent variable capturing the user intention. This intention increases the expressivity of our model, and enables adaptation with generalized policy improvement. We call our proposed method **intention-conditioned flow occupancy models (InFOM)**. Comparing with alternative methods for pre-training, our experiments on $36$ state-based and $4$ image-based benchmark tasks demonstrate that the proposed method achieves $1.8 \times$ median improvement in returns and increases success rates by $36\\%$.

Experience replay enables data-efficient learning from past experiences in online reinforcement learning agents. Traditionally, experiences were sampled uniformly from a replay buffer, regardless of differences in experience-specific learning potential. In an effort to sample more efficiently, researchers introduced Prioritized Experience Replay (PER). In this paper, we propose an extension to PER by introducing a novel measure of temporal difference error reliability. We theoretically show that the resulting transition selection algorithm, Reliability-adjusted Prioritized Experience Replay (ReaPER), enables more efficient learning than PER. We further present empirical results showing that ReaPER outperforms both uniform experience replay and PER across a diverse set of traditional environments including several classic control environments and the Atari-10 benchmark, which approximates the median score across the Atari-57 benchmark within one percent of variance.

Open-ended generation tasks require outputs to satisfy diverse and often implicit task-specific evaluation rubrics. The sheer number of relevant rubrics leads to prohibitively high verification costs and incomplete assessments of a response, making reinforcement learning (RL) post-training with rubric-based rewards difficult to scale. This problem is exacerbated by the fact that often the best way to combine these rubrics into one single reward is also highly prompt-specific. We propose Reinforcement Learning with Adversarial Critic (RLAC), a post-training approach that addresses these challenges via dynamic rubric verification. Our approach employs a large language model (LLM) as a critic that dynamically identifies only the most likely failure modes (e.g., a factual error or unhandled edge case), which are then verified by an external validator to optimize both generator and critic jointly. By training both the generator and the critic, this game enhances the critic's error detection and the generator's output quality while reducing required verifications. Our experiments demonstrate that RLAC improves factual accuracy in text generation and correctness in code generation, while also outperforming exhaustive verification and reward model methods. We show that dynamic critics are more effective than fixed critics, showcasing the potential of RLAC for scaling RL post-training to free-form generation tasks.

Deep Q-learning jointly learns representations and values within monolithic networks, promising beneficial co-adaptation between features and value estimates. Although this architecture has attained substantial success, the coupling between representation and value learning creates instability as representations must constantly adapt to non-stationary value targets, while value estimates depend on these shifting representations. This is compounded by high variance in bootstrapped targets, which causes bias in value estimation in off-policy methods. We introduce Stackelberg Coupled Representation and Reinforcement Learning (SCORER), a framework for value-based RL that views representation and Q-learning as two strategic agents in a hierarchical game. SCORER models the Q-function as the leader, which commits to its strategy by updating less frequently, while the perception network (encoder) acts as the follower, adapting more frequently to learn representations that minimize Bellman error variance given the leader's committed strategy. Through this division of labor, the Q-function minimizes MSBE while perception minimizes its variance, thereby reducing bias accordingly, with asymmetric updates allowing stable co-adaptation, unlike simultaneous parameter updates in monolithic solutions. Our proposed SCORER framework leads to a bi-level optimization problem whose solution is approximated by a two-timescale algorithm that creates an asymmetric learning dynamic between the two players. Extensive experiments on DQN and its variants demonstrate that gains stem from algorithmic insight rather than model complexity.

Sampling efficiency is a key bottleneck in reinforcement learning with verifiable rewards. Existing group-based policy optimization methods, such as GRPO, allocate a fixed number of rollouts for all training prompts. This uniform allocation implicitly treats all prompts as equally informative, and could lead to inefficient computational budget usage and impede training progress. We introduce VIP, a Variance-Informed Predictive allocation strategy that allocates a given rollout budget to the prompts in the incumbent batch to minimize the expected gradient variance of the policy update. At each iteration, VIP uses a lightweight Gaussian process model to predict per-prompt success probabilities based on recent rollouts. These probability predictions are translated into variance estimates, which are then fed into a convex optimization problem to determine the optimal rollout allocations under a hard compute budget constraint. Empirical results show that VIP consistently improves sampling efficiency and achieves higher performance than uniform or heuristic allocation strategies in multiple benchmarks.

Large language models (LLMs) are reshaping the recommender system paradigm by enabling users to express preferences and receive recommendations through conversations. Yet, aligning LLMs to the recommendation task remains challenging: pretrained LLMs often generate out-of-catalog items, violate required output formats, and their ranking quality degrades sharply toward the end of the generated list. To this end, we propose ConvRec-R1, a two-stage framework for end-to-end training of LLM-based conversational recommender systems. In Stage 1, we construct a behavioral-cloning dataset with a Remap-Reflect-Adjust pipeline, which produces high-quality, catalog-grounded demonstrations from powerful blackbox LLMs to warm-start the RL training. In Stage 2, we propose Rank-GRPO, a principled extension of group relative policy optimization (GRPO) tailored to tasks with rank-style outputs. Rank-GRPO treats each rank in the recommendation list as the unit instead of token (too fine-grained) or sequence (too coarse), redefining rewards to remove non-causal credit assignment and introducing a rank-level importance ratio based on the geometric mean of rank-wise token probabilities to stabilize policy updates. Experiments on the on the Reddit-v2 and Redial datasets show that ConvRec-R1 converges faster and achieves higher Recall and NDCG than GRPO-style baselines. Code and datasets are released at https://github.com/yaochenzhu/Rank-GRPO.

Reinforcement Learning with Verifiable Rewards (RLVR) strengthens LLM reasoning but training often oscillates between {entropy collapse} and {entropy explosion}. We trace both hazards to the mean-baseline used in value-free RL (\eg GRPO/DAPO), which improperly penalizes negative-advantage samples under reward outliers. We propose {Quantile Advantage Estimation} (QAE), replacing the mean with a group-wise $K$-quantile baseline. QAE induces a response-level, two-regime gate: on hard queries ($p \le 1{-}K$) it reinforces rare successes, while on easy queries ($p > 1{-}K$) it targets remaining failures. Under first-order softmax updates, we prove {two-sided entropy safety}, giving lower/upper bounds on one-step entropy change that curb explosion and prevent collapse. Empirically, this minimal modification stabilizes entropy, sparsifies credit assignment (with tuned $K$, roughly 80\% of responses receive zero advantage), and yields sustained pass@1 gains on Qwen3-8B/14B-Base across AIME'24/'25 and AMC'23. These results identify {baseline design}—rather than token-level heuristics—as the primary mechanism for scaling RLVR.

Non-prehensile manipulation offers a robust alternative to traditional pick-and-place methods for object repositioning. However, learning such skills with dexterous, multi-fingered hands remains largely unexplored, leaving their potential for stable and efficient manipulation underutilized. Progress has been limited by the lack of large-scale, contact-aware non-prehensile datasets for dexterous hands and the absence of wrist–finger control policies. To bridge these gaps, we present DexMove, a tactile-guided non-prehensile manipulation framework for dexterous hands. DexMove combines a scalable simulation pipeline that generates physically plausible wrist–finger trajectories with a wearable device, which captures multi-finger contact data from human demonstrations using vision-based tactile sensors. Using these data, we train a flow-based policy that enables real-time, synergistic wrist–finger control for robust non-prehensile manipulation of diverse tabletop objects. In real-world experiments, DexMove successfully manipulated six objects of varying shapes and materials, achieving a 77.8\% success rate. Our method outperforms ablated baselines by 36.6\% and improves efficiency by nearly 300\%. Furthermore, the learned policy generalizes to language-conditioned, long-horizon tasks such as object sorting and desktop tidying.

Exploration of dynamic environments with sparse rewards is a significant challenge in Reinforcement Learning, often leading to inefficient exploration and brittle policies. To address this, we introduce a novel graph-based intrinsic reward using Effective Resistance, a metric from spectral graph theory. This reward formulation guides the agent to seek configurations that are directly correlated to successful goal reaching states. We provide theoretical guarantees, proving that our method not only learns a robust policy but also achieves faster convergence by serving as a variance reduction baseline to the standard discounted reward formulation. We perform extensive empirical analysis across several challenging environments to demonstrate that our approach significantly outperforms state-of-the-art baselines, demonstrating improvements of up to 59% in success rate, 56% in timesteps taken to reach the goal, and 4 times more accumulated reward. We augment all of the supporting lemmas and theoretically motivated hyperparameter choices with corresponding experiments.

The real world is inherently non-stationary, with ever-changing factors, such as weather conditions and traffic flows, making it challenging for agents to adapt to varying environmental dynamics. Non-Stationary Reinforcement Learning (NSRL) addresses this challenge by training agents to adapt rapidly to sequences of distinct Markov Decision Processes (MDPs). However, existing NSRL approaches often focus on tasks with regularly evolving patterns, leading to limited adaptability in highly dynamic settings. Inspired by the success of Wavelet analysis in time series modeling, specifically its ability to capture signal trends at multiple scales, we propose WISDOM to leverage wavelet-domain predictive task representations to enhance NSRL. WISDOM captures these multi-scale features in evolving MDP sequences by transforming task representation sequences into the wavelet domain, where wavelet coefficients represent both global trends and fine-grained variations of non-stationary changes. In addition to the auto-regressive modeling commonly employed in time series forecasting, we devise a wavelet temporal difference (TD) update operator to enhance tracking and prediction of MDP evolution. We theoretically prove the convergence of this operator and demonstrate policy improvement with wavelet task representations. Experiments on diverse benchmarks show that WISDOM significantly outperforms existing baselines in both sample efficiency and asymptotic performance, demonstrating its remarkable adaptability in complex environments characterized by non-stationary and stochastically evolving tasks.

Skill-based reinforcement learning (SBRL) enables rapid adaptation in environments with sparse rewards by pretraining a skill-conditioned policy. Effective skill learning requires jointly maximizing both exploration and skill diversity. However, existing methods often face challenges in simultaneously optimizing for these two conflicting objectives. In this work, we propose a new method, Adaptive Multi-objective Projection for balancing Exploration and skill Diversification (AMPED), which explicitly addresses both: during pre-training, a gradient-surgery projection balances the exploration and diversity gradients, and during fine-tuning, a skill selector exploits the learned diversity by choosing skills suited to downstream tasks. Our approach achieves performance that surpasses SBRL baselines across various benchmarks. Through an extensive ablation study, we identify the role of each component and demonstrate that each element in AMPED is contributing to performance. We further provide theoretical evidence that, with a greedy skill selector, greater skill diversity reduces fine-tuning sample complexity. These results highlight the importance of explicitly harmonizing exploration and diversity and demonstrate the effectiveness of AMPED in enabling robust and generalizable skill learning. https://geonwoo.me/amped/

Reinforcement Learning with Verifiable Rewards (RLVR) is a leading approach for tuning language models on mathematical reasoning tasks. However, it remains unclear whether RLVR's gains stem from genuine reasoning improvements or simply from steering the model toward answer formats that already appear in the reference distribution. Inspired by recent evidence \citep{zhao2025echo,yue2025does}, we study this question by optimizing only the first $k$ tokens (e.g. $k=32$) of each solution, generating the remainder of the response from the reference model. We study two methods for prefix optimization, using a naive algorithm that clusters prefixes and selects the best prefix (Prefix Clustering), and a method that optimizes the prefix by finetuning a lightweight adapter model with RL (Prefix-RL). We show that tuning only the first $k$ tokens can significantly improve the accuracy on math, suggesting that at least some of the gains from RL are due to upweighting a preferable solution strategy. Our results suggest that simple prefix optimization methods can provide an efficient alternative to RL, delivering substantial improvements across different models and benchmarks for a tiny fraction of the compute required for standard RL, and that these gains are robust across prefix lengths and random seeds.

Reinforcement learning (RL) for LLM reasoning has rapidly emerged as a prominent research area, marked by a significant surge in related studies on both algorithmic innovations and practical applications. Despite this progress, several critical challenges remain, including the absence of standardized guidelines for applying RL techniques and a fragmented understanding of their underlying mechanisms. In addition, inconsistent experimental settings, variations in training data, and differences in model initialization have led to conflicting conclusions, obscuring the key characteristics of these techniques and creating confusion among practitioners when selecting appropriate techniques. This paper systematically reviews widely adopted RL techniques through rigorous reproductions and isolated evaluations within a unified open-source framework. We analyze the internal mechanisms, applicable scenarios, and core principles of each technique through fine-grained experiments, including datasets of varying difficulty, model sizes, and architectures. Based on these insights, we present clear guidelines for selecting RL techniques tailored to specific setups and provide a reliable roadmap for practitioners navigating the RL for the LLM domain. Finally, we show that a minimalist combination of two techniques can unlock the learning capability of critic-free policies with a vanilla PPO loss. The results demonstrate that our simple combination consistently improves performance, surpassing strategies such as GRPO and DAPO.

We study diffusion-based world models for reinforcement learning, which offer high generative fidelity but face critical efficiency challenges in control. Current methods either require heavyweight models at inference or rely on highly sequential imagination, both of which impose prohibitive computational costs. We propose Horizon Imagination (HI), an on-policy imagination process for discrete stochastic policies that denoises multiple future observations in parallel. HI incorporates a stabilization mechanism and a novel sampling schedule that decouples the denoising budget from the effective horizon over which denoising is applied while also supporting sub-frame budgets. Experiments on Atari 100K and Craftium show that our approach maintains control performance with a sub-frame budget of half the denoising steps and achieves superior generation quality under varied schedules. Code is available at https://github.com/leor-c/horizon-imagination.

In this work, we present a simple yet theoretically motivated improvement to Supervised Fine-Tuning (SFT) for the Large Language Model (LLM), addressing its limited generalization compared to reinforcement learning (RL). Through mathematical analysis, we reveal that standard SFT gradients implicitly encode a problematic reward structure that may severely restrict the generalization capabilities of model compared to RL. To rectify this, we propose Dynamic Fine-Tuning (DFT), stabilizing gradient updates for each token by dynamically rescaling the objective function with the probability of this token. With just a single-line change, the method outperforms standard SFT on multiple difficult benchmarks and base models, from math reasoning to code generation and multi-modal tasks, demonstrating improved generalization. Additionally, \model~achieves competitive results in offline RL settings, providing an effective yet streamlined alternative. By bridging theoretical insights with practical solutions, this work advances the state of SFT. The source code will be available at https://github.com/yongliang-wu/DFT.

Although Reinforcement Learning with Verifiable Rewards (RLVR) has become an essential component for developing advanced reasoning skills in language models, contemporary studies have documented training plateaus after thousands of optimization steps, i.e., notable decreases in performance gains despite increased computational investment. This limitation stems from the sparse exploration patterns inherent in current RLVR practices, where models rely on limited rollouts that often miss critical reasoning paths and fail to provide systematic coverage of the solution space. We present DeepSearch, a framework that integrates Monte Carlo Tree Search (MCTS) directly into RLVR training. In contrast to existing methods that rely on tree search only at inference, DeepSearch embeds structured search into the training loop, enabling systematic exploration and fine-grained credit assignment across reasoning steps. Through training-time exploration, DeepSearch addresses the fundamental bottleneck of insufficient exploration, which leads to diminishing performance gains over prolonged training. Our contributions include: (1) a global frontier selection strategy that prioritizes promising nodes across the search tree, (2) selection with entropy-based guidance that identifies confident paths for supervision, and (3) adaptive replay buffer training with solution caching for efficiency. Experiments on mathematical reasoning benchmarks show that DeepSearch achieves an average accuracy of 62.95% and establishes a new state-of-the-art reasoning model, while using 5.7x fewer GPU hours than extended training approaches. These results highlight the importance of strategic exploration over brute-force scaling and demonstrate the promise of algorithmic innovation for advancing RLVR methodologies. DeepSearch establishes a new direction for scaling reasoning capabilities through systematic search rather than prolonged computation.

Preference-based reinforcement learning (PbRL) enables agent training without explicit reward design by leveraging human feedback. Although various query sampling strategies have been proposed to improve feedback efficiency, many fail to enhance performance because they select queries from outdated experiences with low likelihood under the current policy. Such queries may no longer represent the agent's evolving behavior patterns, reducing the informativeness of human feedback. To address this issue, we propose a policy likelihood-based query sampling and critic-exploited reset (PoLiCER). Our approach uses policy likelihood-based query sampling to ensure that queries remain aligned with the agent’s evolving behavior. However, relying solely on policy-aligned sampling can result in overly localized guidance, leading to overestimation bias, as the model tends to overfit to early feedback experiences. To mitigate this, PoLiCER incorporates a dynamic resetting mechanism that selectively resets the reward estimator and its associated Q-function based on critic outputs. Experimental evaluation across diverse locomotion and robotic manipulation tasks demonstrates that PoLiCER consistently outperforms existing PbRL methods. Our code is available at https://github.com/JongKook-Heo/PoLiCER.

Vision-based reinforcement learning can benefit from object-centric scene representation, which factorizes the visual observation into individual objects and their attributes, such as color, shape, size, and position. While such object-centric representations can extract components that generalize well for various multi-object manipulation tasks, they are prone to issues with occlusions and 3D ambiguity of object properties due to their reliance on single-view 2D image features. Furthermore, the entanglement between object configurations and camera poses complicates the object-centric disentanglement in 3D, leading to poor 3D reasoning by the agent in vision-based reinforcement learning applications. To address the lack of 3D awareness and the object-camera entanglement problem, we propose an enhanced 3D object-centric representation that utilizes multi-view 3D features and enforces more explicit 3D-aware disentanglement. The enhancement is based on the integration of the recent success of multi-view Transformer and the prototypical representation learning among the object-centric representations. The representation, therefore, can stably identify proxies of 3D positions of individual objects along with their semantic and physical properties, exhibiting excellent interpretability and controllability. Then, our proposed block transformer policy effectively performs novel tasks by assembling desired properties adaptive to the new goal states, even when provided with unseen viewpoints at test time. We demonstrate that our 3D-aware block representation is scalable to compose diverse novel scenes and enjoys superior performance in out-of-distribution tasks with multi-object manipulations under both seen and unseen viewpoints compared to existing methods.

Neural networks applied to sequential decision-making tasks typically rely on latent representations of environment states. While environment dynamics dictate how semantic states evolve, the corresponding latent transitions are usually left implicit, creating a potential misalignment between the two. We propose to model latent dynamics explicitly by drawing an analogy between Markov decision process (MDP) trajectories and ordinary differential equation (ODE) flows: in both cases, the current state fully determines its successors. Building on this view, we introduce a neural ODE-based regularization method that enforces latent embeddings to follow consistent ODE flows, thereby aligning representation learning with environment dynamics. Although broadly applicable to deep learning agents, we demonstrate its effectiveness in reinforcement learning by integrating it into Actor-Critic algorithms. Our approach yields major performance gains across various standard Atari benchmarks for A2C and gridworld environments for PPO.

We propose ERA, a new paradigm for entropy-constrained policy via output activation. It guarantees minimum sampling entropy by transforming the outputs of the last layer. Our approach demonstrates broad effectiveness across different domains: 1) for large language models(LLMs), boosting the average score across six benchmarks for Qwen2.5-Math-7B by 11.6%; 2) for continuous control reinforcement learning agents, improving performance by more than 30% over strong baselines such as SAC on the challenging HumanoidBench; 3) for image classification, enhancing ImageNet top-1 accuracy by 0.69% for ResNet-50. These gains are achieved with a computational overhead of less than 7%. Our work validates output activation as a powerful tool for entropy control, opening a new direction for designing simpler and more robust algorithms. Code available at: https://nothingbutbut.github.io/era

Reward design is of great importance for solving complex tasks with reinforcement learning. Recent studies have explored using image-text similarity produced by vision-language models (VLMs) to augment rewards of a task with visual feedback. A common practice linearly adds VLM scores to task or success rewards without explicit shaping, potentially altering the optimal policy. Moreover, such approaches, often relying on single static images, struggle with tasks whose desired behavior involves complex, dynamic motions spanning multiple visually different states. Furthermore, single viewpoints can occlude critical aspects of an agent's behavior. To address these issues, this paper presents Multi-View Video Reward Shaping (MVR), a framework that models the relevance of states regarding the target task using videos captured from multiple viewpoints. MVR leverages video-text similarity from a frozen pre-trained VLM to learn a state relevance function that mitigates the bias towards specific static poses inherent in image-based methods. Additionally, we introduce a state-dependent reward shaping formulation that integrates task-specific rewards and VLM-based guidance, automatically reducing the influence of VLM guidance once the desired motion pattern is achieved. We confirm the efficacy of the proposed framework with extensive experiments on challenging humanoid locomotion tasks from HumanoidBench and manipulation tasks from MetaWorld, verifying the design choices through ablation studies.

Recent advances in applying reinforcement learning (RL) to large language models (LLMs) have led to substantial progress. In particular, a series of remarkable yet often counterintuitive phenomena have been reported in LLMs, exhibiting patterns not typically observed in traditional RL settings. For example, notable claims include that a single training example can match the performance achieved with an entire dataset, that the reward signal does not need to be very accurate, and that training solely with negative samples can match or even surpass sophisticated reward-based methods. However, the precise conditions under which these observations hold—and, critically, when they fail—remain unclear. In this work, we identify a key factor that differentiates RL observations: whether the pretrained model already exhibits strong Model-Task Alignment, as measured by pass@k accuracy on the evaluated task. Through a systematic and comprehensive examination of a series of counterintuitive claims, supported by rigorous experimental validation across different model architectures and task domains, our findings show that while standard RL training remains consistently robust across settings, many of these counterintuitive results arise only when the model and task already exhibit strong model-task alignment. In contrast, these techniques fail to drive substantial learning in more challenging regimes, where standard RL methods remain effective.

Safe policy improvement (SPI) offers theoretical control over policy updates, yet existing guarantees largely concern offline, tabular reinforcement learning (RL). We study SPI in general online settings, when combined with world model and representation learning. We develop a theoretical framework showing that restricting policy updates to a well-defined neighborhood of the current policy ensures monotonic improvement and convergence. This analysis links transition and reward prediction losses to representation quality, yielding online, ''deep'' analogues of classical SPI theorems from the offline RL literature. Building on these results, we introduce DeepSPI, a principled on-policy algorithm that couples local transition and reward losses with regularised policy updates. On the ALE-57 benchmark, DeepSPI matches or exceeds strong baselines, including PPO and DeepMDPs, while retaining theoretical guarantees.

Reinforcement Learning (RL) has demonstrated significant success in optimizing complex control and planning problems. However, scaling RL to real-world applications with multiple, potentially conflicting requirements requires an effective handling of constraints. We propose a novel approach to constraint satisfaction in RL algorithms, focusing on incrementally improving policy safety rather than directly projecting the policy onto a feasible region. We accomplish this by first solving a nonparametric surrogate problem which is guaranteed to contract towards the feasible set, and then cloning that solution into a neural network policy. As a result, our approach improves stability, particularly during early training stages, when the policy lacks knowledge of constraint boundaries. We provide general theoretical results guaranteeing convergence to the safe set for this class of incremental systems. Notably, even the simplest algorithm produced by our theory produces comparable or superior performance when compared to highly tuned constrained RL baselines in challenging constrained environments.

Language models have shown strong performance on mathematical reasoning tasks. Post-training with outcome-based reinforcement learning (RL) can further enhance reasoning but is inefficient because it relies solely on final rewards. Recent credit assignment–based RL methods provide intermediate feedback, yet they often struggle to fairly evaluate each step’s importance, especially in partially correct reasoning chains. We propose Sequential Shapley Value Policy Optimization (SSVPO), a step-level credit assignment framework inspired by multi-agent RL. SSVPO introduces an insertion MDP and Sequential Shapley Values (SSV), which measure each step’s marginal contribution by reordering reasoning steps into alternative chains, ensuring fair credit assignment to all possible steps. By identifying steps with zero credit, SSVPO can shorten reasoning chains to improve training efficiency. We further provide a theoretical proof that SSV fairness to allocate credits and demonstrate that SSV as the new advantage baseline is consistent with Proximal Policy Optimization (PPO). Across 7 benchmarks, SSVPO outperforms state-of-the-art RL methods, both outcome-based (RLOO, GRPO, DAPO) and credit assignment–based (VinePPO, SPO), achieving up to an 11.6\% gain in accuracy, an 18.1\% reduction in token usage, and a 1.6× improvement in reasoning efficiency over vanilla methods. Our findings highlight that SSVPO provides effective step-level credit assignment, advancing post-training LLM reasoning performance while reducing token budgets.

In Reward Learning (ReL), we are given feedback on an unknown target reward, and the goal is to use this information to recover it in order to carry out some downstream application, e.g., planning. When the feedback is not informative enough, the target reward is only partially identifiable, i.e., there exists a set of rewards, called the feasible set, that are equally plausible candidates for the target reward. In these cases, the ReL algorithm might recover a reward function different from the target reward, possibly leading to a failure in the application. In this paper, we introduce a general ReL framework that permits to quantify the drop in "performance" suffered in the considered application because of identifiability issues. Building on this, we propose a robust approach to address the identifiability problem in a principled way, by maximizing the "performance" with respect to the worst-case reward in the feasible set. We then develop Rob-ReL, a ReL algorithm that applies this robust approach to the subset of ReL problems aimed at assessing a preference between two policies, and we provide theoretical guarantees on sample and iteration complexity for Rob-ReL. We conclude with some numerical simulations to illustrate the setting and empirically characterize Rob-ReL.

The success of teams in robotics, nature, and society often depends on the division of labor among diverse specialists; however, a principled explanation for when such diversity surpasses a homogeneous team is still missing. Focusing on multi-agent task allocation problems, we study this question from the perspective of reward design: what kinds of objectives are best suited for heterogeneous teams? We first consider an instantaneous, non-spatial setting where the global reward is built by two generalized aggregation operators: an inner operator that maps the N agents’ effort allocations on individual tasks to a task score, and an outer operator that merges the M task scores into the global team reward. We prove that the curvature of these operators determines whether heterogeneity can increase reward, and that for broad reward families this collapses to a simple convexity test. Next, we ask what incentivizes heterogeneity to emerge when embodied, time-extended agents must learn an effort allocation policy. To study heterogeneity in such settings, we use multi-agent reinforcement learning (MARL) as our computational paradigm, and introduce Heterogeneity Gain Parameter Search (HetGPS), a gradient-based algorithm that optimizes the parameter space of underspecified MARL environments to find scenarios where heterogeneity is advantageous. Across different environments, we show that HetGPS rediscovers the reward regimes predicted by our theory to maximize the advantage of heterogeneity, both validating HetGPS and connecting our theoretical insights to reward design in MARL. Together, these results help us understand when behavioral diversity delivers a measurable benefit.

Discovering insights from a real-world data lake potentially containing unclean, semi-structured, and unstructured data requires a variety of data processing tasks, ranging from extraction and cleaning to integration, analysis, and modeling. This process often also demands domain knowledge and project-specific insight. While AI models have shown remarkable results in reasoning and code generation, their abilities to design and execute complex pipelines that solve these data-lake-to-insight challenges remain unclear. We introduce KramaBench which consists of 104 manually curated and solved challenges spanning 1700 files, 24 data sources, and 6 domains. KramaBench focuses on testing the end-to-end capabilities of AI systems to solve challenges which require automated orchestration of different data tasks. KramaBench also features a comprehensive evaluation framework assessing the pipeline design and individual data task implementation abilities of AI systems. We evaluate 8 LLMs using our single-agent reference framework DS-Guru, alongside both open- and closed-source single- and multi-agent systems, and find that while current agentic systems may handle isolated data-science tasks and generate plausible draft pipelines, they struggle with producing working end-to-end pipelines. On KramaBench, the best system reaches only 55% end-to-end accuracy in the full data-lake setting. Even with perfect retrieval, the accuracy tops out at 62%. Leading LLMs can identify up to 42% of important data tasks but can only fully implement 20% of individual data tasks.

In cooperative multi-agent reinforcement learning, agents often face scalability challenges due to the exponential growth of the joint action and observation spaces. Inspired by the structure of human teams, we explore subteam-based coordination, where agents are partitioned into fully correlated subgroups with limited inter-group interaction. We formalize this structure using Bayesian networks and propose a class of correlated joint policies induced by directed acyclic graphs . Theoretically, we prove that regularized policy gradient ascent converges to near-optimal policies under a decomposability condition of the environment. Empirically, we introduce a heuristic for dynamically constructing context-aware subteams with limited dependency budgets, and demonstrate that our method outperforms standard baselines across multiple benchmark environments.

Test-time reasoning significantly enhances pre-trained AI agents’ performance. However, it requires an explicit environment model, often unavailable or overly complex in real-world scenarios. While MuZero enables effective model learning for search in perfect information games, extending this paradigm to imperfect information games presents substantial challenges due to more nuanced look-ahead reasoning techniques and large number of states relevant for individual decisions. This paper introduces an algorithm LAMIR that learns an abstracted model of an imperfect information game directly from the agent-environment interaction. During test time, this trained model is used to perform look-ahead reasoning. The learned abstraction limits the size of each subgame to a manageable size, making theoretically principled look-ahead reasoning tractable even in games where previous methods could not scale. We empirically demonstrate that with sufficient capacity, LAMIR learns the exact underlying game structure, and with limited capacity, it still learns a valuable abstraction, which improves game playing performance of the pre-trained agents even in large games.

Autonomous web agents solve complex browsing tasks, yet existing benchmarks measure only whether an agent finishes a task, ignoring whether it does so safely or in a way enterprises can trust. To integrate these agents into critical workflows, safety and trustworthiness (ST) are prerequisite conditions for adoption. We introduce ST-WebAgentBench, a configurable and extensible framework designed as a first step toward enterprise-grade evaluation. Each of its 375 tasks carries one or more ST policies (3,057 in total), concise rules encoding constraints, and is scored along six orthogonal dimensions (e.g., user consent, robustness). Tasks span three difficulty tiers for fine-grained capability profiling, and a “Modality Challenge” disentangles vision-only from DOM-only information retrieval, isolating the contribution of each perceptual modality to agent failures. Beyond raw task success, we propose the Completion Under Policy (CuP) metric, which credits only completions that respect all applicable policies, and the Risk Ratio, which quantifies ST breaches across dimensions. Evaluating three open state-of-the-art agents shows their average CuP is less than two-thirds of their nominal completion rate, revealing substantial safety gaps. To support growth and adaptation to new domains, ST-WebAgentBench provides modular code and extensible templates that enable new workflows to be incorporated with minimal effort, offering a practical foundation for advancing trustworthy web agents at scale.

Outcome-driven reinforcement learning has advanced reasoning in large language models (LLMs), but prevailing tool-augmented approaches train a single, monolithic policy that interleaves thoughts and tool calls under full context; this scales poorly with long horizons and diverse tools and generalizes weakly to new scenarios. Agentic systems offer a promising alternative by decomposing work across specialized modules, yet most remain training-free or rely on offline training decoupled from the live dynamics of multi-turn interaction. We introduce AgentFlow, a trainable, in-the-flow agentic framework that coordinates four modules (planner, executor, verifier, generator) through an evolving memory and directly optimizes its planner inside the multi-turn loop. To train on-policy in live environments, we propose Flow-based Group Refined Policy Optimization (Flow-GRPO), which tackles long-horizon, sparse-reward credit assignment by converting multi-turn optimization into a sequence of tractable single-turn policy updates. It broadcasts a single, verifiable trajectory-level outcome to every turn to align local planner decisions with global success and stabilizes learning with group-normalized advantages. Across ten benchmarks, AgentFlow with a 7B-scale backbone outperforms top-performing baselines with average accuracy gains of 14.9% on search, 14.0% on agentic, 14.5% on mathematical, and 4.1% on scientific tasks, even surpassing larger proprietary models like GPT-4o. Further analyses confirm the benefits of in-the-flow optimization, showing improved planning, enhanced tool-calling reliability, and positive scaling with model size and reasoning turns.

Existing reinforcement learning (RL) methods struggle with complex dynamical systems that demand interactions at high frequencies or irregular time intervals. Continuous-time RL (CTRL) has emerged as a promising alternative by replacing discrete-time Bellman recursion with differentiable value functions defined as viscosity solutions of the Hamilton–Jacobi–Bellman (HJB) equation. While CTRL has shown promise, its applications have been largely limited to the single-agent domain. This limitation stems from two key challenges: (i) conventional methods for solving HJB equations suffer from the curse of dimensionality (CoD), making them intractable in high-dimensional systems; and (ii) even with learning-based approaches to alleviate the CoD, accurately approximating centralized value functions in multi-agent settings remains difficult, which in turn destabilizes policy training. In this paper, we propose a CT-MARL framework that uses physics-informed neural networks (PINNs) to approximate HJB-based value functions at scale. To ensure the value is consistent with its differential structure, we align value learning with value-gradient learning by introducing a Value Gradient Iteration (VGI) module that iteratively refines value gradients along trajectories. This improves gradient accuracy, in turn yielding more precise value approximations and stronger policy learning. We evaluate our method using continuous‑time variants of standard benchmarks, including multi‑agent particle environment (MPE) and multi‑agent MuJoCo. Our results demonstrate that our approach consistently outperforms existing continuous‑time RL baselines and scales to complex cooperative multi-agent dynamics. Code is available at https://github.com/Wangxuefeng1024/Continuous-Time-Value-Iteration-for-Multi-Agent-Reinforcement-Learning.git.

Solving math problems through verifiable languages such as Lean has significantly impacted both the mathematics and computer science communities. Current state-of-the-art models are often trained with expensive online Reinforcement Learning (RL) or expert iteration. However, these approaches rely on fixed problem sets, which causes inefficient training and limits the model to tackle complex problems. To overcome these limitations, we propose GAR: Generative Adversarial Reinforcement learning, a comprehensive RL training framework that jointly trains the problem composer and solver in an adversarial loop. GAR introduces an implicit curriculum learning mechanism, which aligns task difficulty with the prover's evolving capability. It thereby improves the training efficiency and enables stronger performance of proving advanced theorems. Experiments show that with GAR training, Goedel-Prover-V2-8B and DeepSeek-Prover-V2-7B achieve an average relative improvement in pass@32 of 4.20% on MiniF2F-Test benchmark, while DeepSeek-Prover-V2's pass@32 on ProofNet-Test increases from 22.58% to 25.81%. Beyond formal proving, GAR establishes a general RL paradigm for co-evolution of problem generation and solving under verifiable environments. The training code for this paper is open-sourced in https://github.com/RickySkywalker/GAR-Official

We consider a Multi-agent Reinforcement Learning (MARL) setting, in which an attacker can arbitrarily corrupt any subset of up to $k$ out of $n$ agents at deployment. Our goal is to design agents that are robust against such an attack, by accounting for the presence of corrupted agents at test time. To that end, we introduce a novel solution concept, the Adversarially Robust Nash Equilibrium (ARNEQ), and provide theoretical proof of its existence in general-sum Markov games. Furthermore, we introduce a proof-of-concept model-based approach to computing it and theoretically prove its convergence under standard assumptions. We also present a practical approach called Adversarially Robust Training (ART), an independent learning algorithm based on stochastic gradient descent ascent. Our experiments in both cooperative and mixed cooperative-competitive environments demonstrate ART's effectiveness and practical value in enhancing MARL resilience against adversarial behavior.

Variance-dependent regret bounds for linear contextual bandits, which improve upon the classical $\tilde{O}(d\sqrt{K})$ regret bound to $\tilde{O}(d\sqrt{\sum_{k=1}^K\sigma_k^2})$, where $d$ is the context dimension, $K$ is the number of rounds, and $\sigma^2_k$ is the noise variance in round $k$, has been widely studied in recent years. However, most existing works focus on the regret upper bounds instead of lower bounds. To our knowledge, the only lower bound is from Jia et al. (2024), which proved that for any eluder dimension $d_{\textbf{elu}}$ and total variance budget $\Lambda$, there exists an instance with $\sum_{k=1}^K\sigma_k^2\leq \Lambda$ for which any algorithm incurs a variance-dependent lower bound of $\Omega(\sqrt{d_{\textbf{elu}}\Lambda})$. However, this lower bound has a $\sqrt{d}$ gap with existing upper bounds. Moreover, it only considers a fixed total variance budget $\Lambda$ and does not apply to a general variance sequence $\{\sigma_1^2,\ldots,\sigma_K^2\}$. In this paper, to overcome the limitations of Jia et al. (2024), we consider the general variance sequence under two settings. For a prefixed sequence, where the entire variance sequence is revealed to the learner at the beginning of the learning process, we establish a variance-dependent lower bound of $\Omega(d \sqrt{\sum_{k=1}^K\sigma_k^2 }/\log K)$ for linear contextual bandits. For an adaptive sequence, where an adversary can generate the variance $\sigma_k^2$ in each round $k$ based on historical observations, we show that when the adversary must generate $\sigma_k^2$ before observing the decision set $D_k$, a similar lower bound of $\Omega(d\sqrt{ \sum_{k=1}^K\sigma_k^2} /\log^6(dK))$ holds. In both settings, our results match the upper bounds of the SAVE algorithm (Zhao et al. 2023) up to logarithmic factors. Furthermore, if the adversary can generate the variance $\sigma_k$ after observing the decision set $D_k$, we construct a counter-example showing that it is impossible to construct a variance-dependent lower bound if the adversary properly selects variances in collaboration with the learner. Our lower bound proofs use a novel peeling technique that groups rounds by variance magnitude. For each group, we construct separate instances and assign the learner distinct decision sets. We believe this proof technique may be of independent interest.

From a first-principles perspective, it may seem odd that the strongest results in foundation model fine-tuning (FT) are achieved via a relatively complex, two-stage training procedure. Specifically, one first trains a reward model (RM) on some dataset (e.g., human preferences) before using it to provide online feedback as part of a downstream reinforcement learning (RL) procedure, rather than directly optimizing the policy parameters on said dataset via offline maximum likelihood estimation. In fact, from an information-theoretic perspective, we can only lose information via passing through a reward model and cannot create any new information via on-policy sampling. To explain this discrepancy, we scrutinize several hypotheses on the value of RL in FT through both theoretical and empirical lenses. Of the hypotheses considered, we find the most support for the explanation that on problems with a generation-verification gap, (1) it is relatively easy to learn the relatively simple RM (verifier) from the preference data. Then, (2) the downstream RL procedure only returns policies (generators) that are optimal for such relatively simple verifiers. Thus, end-to-end, two-stage online FT only has to search over a reduced subset of the full space of policies, requiring less data than offline FT.

We introduce ComputerRL, a framework for autonomous desktop intelligence that enables agents to operate complex digital workspaces skillfully. ComputerRL features the API-GUI paradigm, which unifies programmatic API calls and direct GUI interaction to address the inherent mismatch between machine agents and human-centric desktop environments. Scaling end-to-end RL training is crucial for improvement and generalization across diverse desktop tasks; however, it remains challenging due to environmental inefficiency and instability during extended training. To support scalable and robust training, we develop a distributed RL infrastructure capable of orchestrating thousands of parallel virtual desktop environments to accelerate large-scale online RL. Furthermore, we propose Entropulse, a training strategy that alternates reinforcement learning with supervised fine-tuning, effectively mitigating entropy collapse during extended training runs. We employ ComputerRL on open models GLM-4-9B-0414 and GLM-4.1V-9B-Thinking, and evaluate them on the OSWorld benchmark. The GLM-ComputerRL-9B achieves a new state-of-the-art accuracy of 48.9%, demonstrating significant improvements for general agents in desktop automation. Our code is available at https://github.com/THUDM/ComputerRL.

At the boundary between the known and the unknown, an agent inevitably confronts the dilemma of whether to explore or to exploit. Epistemic uncertainty reflects such boundaries, representing systematic uncertainty due to limited knowledge. In this paper, we propose a Bayesian reinforcement learning (RL) algorithm, $\texttt{EUBRL}$, which leverages epistemic guidance to achieve principled exploration. This guidance adaptively reduces per-step regret arising from estimation errors. We establish nearly minimax-optimal regret and sample complexity guarantees for a class of sufficiently expressive priors in infinite-horizon discounted MDPs. Empirically, we evaluate $\texttt{EUBRL}$ on tasks characterized by sparse rewards, long horizons, and stochasticity. Results demonstrate that $\texttt{EUBRL}$ achieves superior sample efficiency, scalability, and consistency.

The rapid pace at which new large language models (LLMs) appear, and older ones become obsolete, forces providers to manage a streaming inventory under a strict concurrency cap and per-query cost budgets. We cast this as an online decision problem that couples *stage-wise deployment* (at fixed maintenance windows) with *per-query routing* among live models. We introduce *StageRoute*, a hierarchical algorithm that (i) optimistically selects up to $M_{\max}$ models for the next stage using reward upper-confidence and cost lower-confidence bounds, and (ii) routes each incoming query by solving a budget- and throughput-constrained bandit subproblem over the deployed set. We prove a regret of $\tilde{\mathcal{O}}(T^{2/3})$ with a matching lower bound, establishing near-optimality, and validate the theory empirically: *StageRoute* tracks a strong oracle under tight budgets across diverse workloads.

Building general-purpose graphical user interface (GUI) agents has become increasingly promising with the progress in vision language models. However, developing effective mobile GUI agents with reinforcement learning (RL) remains challenging due to the heavy-tailed distribution of task difficulty and the inefficiency of large-scale environment sampling. We present an online agentic reinforcement learning framework MOBILERL to enhance GUI agents in mobile environments. Its core component is the Difficulty-Adaptive GRPO (ADAGRPO) algorithm. In ADAGRPO, we design difficulty-adaptive positive replay and failure curriculum filtering to adapt the model to different task difficulties. We introduce the shortest-path reward adjustment strategy to reshape rewards concerning the task length in multi-turn agentic tasks. Those strategies jointly stabilize RL training, improve sample efficiency, and generate strong performance across diverse mobile apps and tasks. We apply MOBILERL to two open models (Qwen2.5-VL-7B-Instruct and GLM-4.1V-9B-Base). The resultant MOBILERL-9B model achieves state-of-the-art results in terms of success rates on both AndroidWorld (80.2%) and Android-Lab (53.6%). The MOBILERL is open-sourced at https://github.com/THUDM/MobileRL.

Deep neural networks that incorporate classic reinforcement learning methods, such as value iteration, into their structure significantly outperform randomly structured networks in learning and generalization. These networks, however, are mostly limited to environments with no or very low uncertainty and do not extend well to partially observable environments. In this paper, we propose a new planning module architecture, the VI$^2$N (Value Iteration with Value of Information Network), that learns to act in novel environments with high perceptual ambiguity. This architecture over-emphasizes reducing uncertainty before exploiting the reward. VI$^2$N can also utilize factorization in environments with mixed observability to decrease the computational complexity of calculating the policy and to facilitate learning. Tested on a range of grid-based navigation tasks, each containing various types of environments with different degrees of observability, our network outperforms other deep architectures. Moreover, VI$^2$N generates interpretable cognitive maps highlighting both rewarding and informative locations. These maps highlight the key states the agent must visit to achieve its goal.

Search agents have achieved significant advancements in enabling intelligent information retrieval and decision-making within interactive environments. Although reinforcement learning has been employed to train agentic models capable of more dynamic interactive retrieval, existing methods are limited by shallow tool-use depth and the accumulation of errors over multiple iterative interactions. In this paper, we present WebSeer, a more intelligent search agent trained via reinforcement learning enhanced with a self-reflection mechanism. Specifically, we construct a large dataset annotated with reflection patterns and design a two-stage training framework that unifies cold start and reinforcement learning within the self-reflection paradigm for real-world web-based environments, which enables the model to generate longer and more reflective tool-use trajectories. Our approach substantially extends tool-use chains and improves answer accuracy. Using a single 14B model, we achieve state-of-the-art results on HotpotQA and SimpleQA, with accuracies of 72.3\% and 90.0\%, respectively, and demonstrate strong generalization to out-of-distribution datasets.

Vision–Language–Action (VLA) systems follow instructions to execute multi-step tasks in multimodal environments. Recent VLA approaches typically rely on post-hoc correction mechanisms or operate under fixed task decompositions and alignment schemes. However, once an intermediate step is mis-specified, local errors propagate through subsequent steps and eventually accumulate into cascading failures. To mitigate this compounding effect, we propose Predictive Alignment and Planning Architecture (ReCAPA), a framework that uses prediction and contrast to adjust deviations across three levels: actions, subgoals, and trajectories. Semantic alignment is enforced at all levels using a Sinkhorn-based module and a Score-field module. The predictive correction and alignment, jointly updates the action-generator in the training phase, enabling it to adjust fine-grained steps to remain aligned with the overall intent. We further introduce two new metrics to quantify error propagation and recovery processes in tasks, capturing how mistakes spread and fade over long-horizon execution. Experiments show that ReCAPA achieves competitive results on embodied agent benchmarks such as VisualAgentBench, MineDojo, and AI2-THOR, outperforming strong proprietary and open-source Large Language Model (LLM) baselines.

Multi-step reasoning tasks like mathematical problem solving are vulnerable to cascading failures, where a single incorrect step leads to complete solution breakdown. Current LLM routing methods assign entire queries to one model, treating all reasoning steps as equal. We propose TRIM (Targeted routing in multi-step reasoning tasks), which routes only critical steps$\unicode{x2013}$those likely to derail the solution$\unicode{x2013}$to larger models while letting smaller models handle routine continuations. Our key insight is that targeted step-level interventions can fundamentally transform inference efficiency by confining expensive calls to precisely those steps where stronger models prevent cascading errors. TRIM operates at the step-level: it uses process reward models to identify erroneous steps and makes routing decisions based on step-level uncertainty and budget constraints. We develop several routing strategies within TRIM, ranging from a simple threshold-based policy to more expressive policies that reason about long-horizon accuracy-cost trade-offs and uncertainty in step-level correctness estimates. On MATH-500, even the simplest thresholding strategy surpasses prior routing methods with 5x higher cost efficiency, while more advanced policies match the strong, expensive model's performance using 80% fewer expensive model tokens. On harder benchmarks such as AIME, TRIM achieves up to 6x higher cost efficiency. All methods generalize effectively across math reasoning tasks, demonstrating that step-level difficulty represents fundamental characteristics of reasoning.

We introduce Gaia2, a benchmark for evaluating large language model agents in realistic, asynchronous environments. Unlike prior static or synchronous evaluations, Gaia2 introduces scenarios where environments evolve independently of agent actions, requiring agents to operate under temporal constraints, adapt to noisy and dynamic events, resolve ambiguity, and collaborate with other agents. Each scenario is paired with a write-action verifier, enabling fine-grained, action-level evaluation and making Gaia2 directly usable for reinforcement learning from verifiable rewards. Our evaluation of state-of-the-art proprietary and open-source models shows that no model dominates across capabilities: GPT-5 (high) reaches the strongest overall score of 42% pass@1 but fails on time-sensitive tasks, Claude-4 Sonnet trades accuracy and speed for cost, Kimi-K2 leads among open-source models with 21% pass@1. These results highlight fundamental trade-offs between reasoning, efficiency, robustness, and expose challenges in closing the “sim2real” gap. Gaia2 is built on a consumer environment with the open-source Agents Research Environments platform and designed to be easy to extend. By releasing Gaia2 alongside the foundational ARE framework, we aim to provide the community with a flexible infrastructure for developing, benchmarking, and training the next generation of practical agent systems.

Reinforcement learning (RL) is widely used to post-train large language models for tasks such as mathematical reasoning and coding. However, the convergence of RL training remains sensitive to batching and prompt selection strategies. We investigate the factors that affect convergence, including batch size and prompt difficulty. Through large-scale experiments across multiple models and datasets, we show that there exists an optimal batch size that balances generation time and gradient quality, and that prompts of intermediate difficulty (where the model has roughly a 50\% chance of success) are the most sample-efficient for model convergence. Motivated by these findings, we propose Prompt Curriculum Learning (PCL), a lightweight algorithm that selects intermediate-difficulty prompts using a learned value model. PCL avoids costly rollouts and efficiently guides training by focusing on the most informative samples. Empirically, PCL either achieves the highest performance or requires significantly less training time to reach comparable performance across a suite of benchmarks. Compared to using rollouts to filter, PCL is $12.1\times$ and $16.9\times$ faster on identifying intermediate-difficulty prompts when training on MATH and DeepScaleR respectively.

Tool use represents a critical capability for AI agents, with recent advances focusing on leveraging reinforcement learning (RL) for test-time scaling to achieve better performance through more deliberate reasoning. However, there are some key challenges in current RL-based scaling approaches: (a) direct RL training often struggles to scale up thinking length sufficiently to solve complex problems, and (b) scaled-up models tend to overthink simpler problems, resulting in substantial token inefficiency. To address these challenges, we propose a novel training paradigm that first employs warm-up supervised fine-tuning to help models distinguish between simple and complex problems, followed by RL that enable models to automatically determine appropriate reasoning trajectories. Furthermore, to tackle the issue of automatic thinking-length scaling, we discover that entropy-based optimization objectives effectively maintain model diversity while successfully unlocking the model's scaling capabilities. Based on this insight, we introduce an entropy-based long-short reasoning fusion RL strategy. Our experiments on three benchmarks demonstrate that model successfully achieves auto-scaling for efficient tool use, achieving significant 9.8\% accuracy improvements while reducing computational overhead by ~81\%.

Enhancing the complex reasoning capabilities of Large Language Models (LLMs) attracts widespread attention. While reinforcement learning (RL) has shown superior performance for improving complex reasoning, its impact on cross-lingual generalization compared to Supervised Fine-Tuning (SFT) remains unexplored. We present the first systematic investigation into cross-lingual reasoning generalization of RL and SFT. Using Qwen2.5-3B-Base as our foundation model, we conduct experiments on diverse multilingual reasoning benchmarks, including math reasoning, commonsense reasoning, and scientific reasoning. Our investigation yields two significant findings: (1) Tuning with RL not only achieves higher accuracy but also demonstrates substantially stronger cross-lingual generalization capabilities compared to SFT. (2) RL training on non-English data yields better overall performance and generalization than training on English data, which is not observed with SFT. Furthermore, through comprehensive mechanistic analyses, we explore the underlying factors of RL's superiority and generalization across languages. Our results provide compelling evidence that RL enables the model with more robust reasoning strategies, offering crucial guidance for more equitable and effective multilingual reasoning.

Large Language Models (LLMs) have recently demonstrated strong potential in generating ‘believable human-like’ behavior in web environments. Prior work has explored augmenting training data with LLM-synthesized rationales and applying supervised fine-tuning (SFT) to enhance reasoning ability, which in turn can improve downstream action prediction. However, the performance of such approaches remains inherently bounded by the reasoning capabilities of the model used to generate the rationales. In this paper, we introduce Shop-R1, a novel reinforcement learning (RL) framework aimed at enhancing the reasoning ability of LLMs for simulation of real human behavior in online shopping environments. Specifically, Shop-R1 decomposes the human behavior simulation task into two stages: rationale generation and action prediction, each guided by distinct reward signals. For rationale generation, we leverage internal model signals (e.g., logit distributions) to guide the reasoning process in a self-supervised manner. For action prediction, we propose a hierarchical reward structure with difficulty-aware scaling to prevent reward hacking and enable fine-grained reward assignment. This design evaluates both high-level action types and the correctness of fine-grained sub-action details (attributes and values), rewarding outputs proportionally to their difficulty. Experimental results show that our method achieves a relative improvement of over 65% compared to the baseline. The project page is available at https://damon-demon.github.io/shop-r1.html.

Large Reasoning Models (LRMs) have demonstrated remarkable performance on complex reasoning tasks by employing test-time scaling. However, they often generate over-long chains-of-thought that, driven by substantial reflections such as repetitive self-questioning and circular reasoning, lead to high token consumption, substantial computational overhead, and increased latency without improving accuracy, particularly in smaller models. Our observation reveals that increasing problem complexity induces more excessive and unnecessary reflection, which in turn reduces accuracy and increases token overhead. To address this challenge, we propose Adaptive Reflection and Length Coordinated Penalty (ARLCP), a novel reinforcement learning framework designed to dynamically balance reasoning efficiency and solution accuracy. ARLCP introduces two key innovations: (1) a reflection penalty that adaptively curtails unnecessary reflective steps while preserving essential reasoning, and (2) a length penalty calibrated to the estimated complexity of the problem. By coordinating these penalties, ARLCP encourages the model to generate more concise and effective reasoning paths. We evaluate our method on five mathematical reasoning benchmarks using DeepSeek-R1-Distill-Qwen-1.5B and DeepSeek-R1-Distill-Qwen-7B models. Experimental results show that ARLCP achieves a superior efficiency-accuracy trade-off compared to existing approaches. For the 1.5B model, it reduces the average response length by 53.1% while simultaneously improving accuracy by 5.8%. For the 7B model, it achieves a 35.0% reduction in length with a 2.7% accuracy gain. The code is released at https://github.com/ZeweiYu1/ARLCP.

The problem of learning how to reach goals in an environment has been a long- standing challenge in for AI researchers. Effective goal-conditioned reinforcement learning (GCRL) methods promise to enable reaching distant goals without task- specific rewards by stitching together past experiences of different complexity. Mathematically, there is a duality between the notion of optimal goal-reaching value functions (the likelihood of success at reaching a goal) and temporal dis- tances (transit times states). Recent works have exploited this property by learning quasimetric distance representations that stitch long-horizon behaviors using the in- ductive bias of their architecture. These methods have shown promise in simulated benchmarks, reducing value learning to a shortest-path problem. But quasimet- ric, and more generally, goal-conditioned RL methods still struggle in complex environments with stochasticity and high-dimensional (visual) observations. There is a fundamental tension between the local dynamic programming (TD backups, temporal distances) that enables optimal shortest-path reasoning in theory and the statistical global MC updates (multistep returns, suboptimal in theory). We show how these approaches can be integrated into a practical GCRL method that fits a quasimetric distance using a multistep Monte-Carlo return. We show our method outperforms existing GCRL methods on long-horizon simulated tasks with up to 4000 steps, even with visual observations. We also demonstrate that our method can enable stitching in the real-world robotic manipulation domain (Bridge setup). Our approach is the first end-to-end GCRL method that enables multistep stitching in this real-world manipulation domain from an unlabeled offline dataset of visual observations.

Designing robust reinforcement learning (RL) agents in the presence of imperfect reward signals remains a core challenge. In practice, agents are often trained with proxy rewards that only approximate the true objective, leaving them vulnerable to reward hacking, where high proxy returns arise from unintended or exploitative behaviors. Recent work formalizes this issue using r-correlation between proxy and true rewards, but existing methods like occupancy-regularized policy optimization (ORPO) optimize against a fixed proxy and do not provide strong guarantees against broader classes of correlated proxies. In this work, we formulate reward hacking as a robust policy optimization problem over the space of all r-correlated proxy rewards. We derive a tractable max-min formulation, where the agent maximizes performance under the worst-case proxy consistent with the correlation constraint. We further show that when the reward is a linear function of known features, our approach can be adapted to incorporate this prior knowledge, yielding both improved policies and interpretable worst-case rewards. Experiments across several environments show that our algorithms consistently outperform ORPO in worst-case returns, and offer improved robustness and stability across different levels of proxy–true reward correlation. These results show that our approach provides both robustness and transparency in settings where reward design is inherently uncertain.

A major bottleneck in scientific discovery involves narrowing an exponentially large set of objects, such as proteins or molecules, to a small set of promising candidates with desirable properties. While this process can rely on expert knowledge, recent methods leverage reinforcement learning (RL) guided by a proxy reward function to enable this filtering. By employing various forms of entropy regularization, these methods aim to learn samplers that generate diverse candidates that are highly rated by the proxy function. In this work, we make two main contributions. First, we show that these methods are liable to generate overly diverse, suboptimal candidates in large search spaces. To address this issue, we introduce a novel unified operator that combines several regularized RL operators into a general framework that better targets peakier sampling distributions. Secondly, we offer a novel, robust RL perspective of this filtering process. The regularization can be interpreted as robustness to a compositional form of uncertainty in the proxy function (i.e., the true evaluation of a candidate differs from the proxy's evaluation). Our analysis leads us to a novel, easy-to-use algorithm we name trajectory general mellowmax (TGM): we show it identifies higher quality, diverse candidates than baselines in both synthetic and real-world tasks.

A burgeoning area within reinforcement learning (RL) is the design of sequential decision-making agents centered around large language models (LLMs). While autonomous decision-making agents powered by modern LLMs could facilitate numerous real-world applications, such successes demand agents that are capable of data-efficient RL. One key obstacle to achieving data efficiency in RL is exploration, a challenge that we demonstrate many recent proposals for LLM agent designs struggle to contend with. Meanwhile, classic algorithms from the RL literature known to gracefully address exploration require technical machinery that can be challenging to operationalize in purely natural language settings. In this work, rather than relying on finetuning or in-context learning to coax LLMs into implicitly imitating a RL algorithm, we illustrate how LLMs can be used to explicitly implement an existing RL algorithm (Posterior Sampling for Reinforcement Learning) whose capacity for statistically-efficient exploration is already well-studied. We offer empirical results demonstrating how our LLM-based implementation of a known, data-efficient RL algorithm can be considerably more effective in natural language tasks that demand prudent exploration.

Execution-based feedback like unit testing is widely used in the development of coding agents through test-time scaling (TTS) and reinforcement learning (RL). This paradigm requires scalable and reliable collection of unit test cases to provide accurate feedback, and the resulting feedback is often sparse and cannot effectively distinguish between trajectories that are both successful or both unsuccessful. In contrast, execution-free feedback from reward models can provide more fine-grained signals without depending on unit test cases. Despite this potential, execution-free feedback for realistic software engineering (SWE) agents remains underexplored. Aiming to develop versatile reward models that are effective across TTS and RL, however, we observe that two verifiers with nearly identical TTS performance can nevertheless yield very different results in RL. Intuitively, TTS primarily reflects the model’s ability to select the best trajectory, but this ability does not necessarily generalize to RL. To address this limitation, we identify two additional aspects that are crucial for RL training: classification accuracy and calibration. We then conduct comprehensive controlled experiments to investigate how to train a robust reward model that performs well across these metrics. In particular, we analyze the impact of various factors such as training data scale, policy mixtures, and data source composition. Guided by these investigations, we introduce SWE-RM, an accurate and robust reward model adopting a mixture-of-experts architecture with 30B total parameters and 3B activated during inference. SWE-RM substantially improves SWE agents on both TTS and RL performance. For example, it increases the accuracy of Qwen3-Coder-Flash from 51.6% to 62.0%, and Qwen3-Coder-Max from 67.0% to 74.6% on SWE-Bench Verified using TTS, achieving new state-of-the-art performance among open-source models. On RL training, SWE-RM lifts the resolve rate of execution-based counterparts by 3 absolute points on SWE-Bench Verified.

We propose a sample-efficient Q-learning algorithm for reinforcement learning with the Conditional Value-at-Risk (CVaR) objective. Our algorithm is built upon predictive tail value function, a novel formulation of risk-sensitive action value, that admits a recursive structure as in the conventional risk-neutral Bellman equation. This structure enables the Q-learning algorithm to utilize the entire set of sample trajectories rather than relying only on worst-case outcomes, enhancing the sample efficiency. We further derive a Bellman optimality equation and a policy improvement theorem, which provide theoretical foundations of our algorithm and remedy inconsistencies that have existed in the literature. Empirical results demonstrate that our method consistently improves CVaR performance while maintaining stable and interpretable learning dynamics.

Reward models (RMs) play a critical role in aligning AI behaviors with human preferences, yet they face two fundamental challenges: (1) Modality Imbalance, where most RMs are mainly focused on text and image modalities, offering limited support for video, audio, and other modalities; and (2) Preference Rigidity, where training on fixed binary preference pairs fails to capture the complexity and diversity of personalized preferences. To address the above challenges, we propose Omni-Reward, a step toward generalist omni-modal reward modeling with support for free-form preferences, consisting of: (1) Evaluation: We introduce Omni-RewardBench, the first omni-modal RM benchmark with free-form preferences, covering nine tasks across five modalities including text, image, video, audio, and 3D; (2) Data: We construct Omni-RewardData, a multimodal preference dataset comprising 248K general preference pairs and 69K instruction-tuning pairs for training generalist omni-modal RMs; (3) Model: We propose Omni-RewardModel, which includes both discriminative and generative RMs, and achieves strong performance on Omni-RewardBench as well as other widely used reward modeling benchmarks.

Off-policy reinforcement learning (RL) for large language models (LLMs) is attracting growing interest, driven by practical constraints in real-world applications, the complexity of LLM-RL infrastructure, and the need for further innovations of RL methodologies. While classic REINFORCE and its modern variants like Group Relative Policy Optimization (GRPO) are typically regarded as on-policy algorithms with limited tolerance of off-policyness, we present in this work a first-principles derivation for group-relative REINFORCE — a REINFORCE variant that uses the within-group mean reward as the baseline for advantage calculation — without assuming a specific training data distribution, showing that it admits a native off-policy interpretation. This perspective yields two general principles for adapting REINFORCE to truly off-policy settings: regularizing policy updates, and actively shaping the data distribution. Our analysis demystifies some myths about the roles of importance sampling and clipping in GRPO, unifies and reinterprets two recent algorithms — Online Policy Mirror Descent and Asymmetric REINFORCE — as regularized forms of the REINFORCE loss, and offers theoretical justification for seemingly heuristic data-weighting strategies. Our findings lead to actionable insights that are validated with extensive empirical studies, and open up new opportunities for principled algorithm design in off-policy RL for LLMs. Source code for this work is available at https://github.com/agentscope-ai/Trinity-RFT/tree/main/examples/rec_gsm8k.

Reinforcement learning (RL) has enabled the training of Large Language Model (LLM) agents to interact with the environment and to solve multi-turn longhorizon tasks. However, the RL-trained agents often struggle in tasks that require active exploration and fail to efficiently adapt from trial-and-error experiences. In this paper, we present LaMer, a general Meta-RL framework that enables LLM agents to actively explore and learn from the environment feedback at test time. LaMer consists of two key components: (i) a cross-episode training framework to encourage exploration and long term rewards optimization; and (ii) in-context policy adaptation via reflection, allowing the agent to adapt their policy from task feedback signal without gradient update. Experiments across diverse environments show that LaMer significantly improves performance over RL baselines, with 11\%, 14\%, and 19\% performance gains on Sokoban, MineSweeper and Webshop, respectively. Moreover, LaMer also demonstrates better generalization to more challenging or previously unseen tasks compared to the RL-trained agents. Overall, our results demonstrate that meta-reinforcement learning provides a principled approach to induce exploration in language agents, enabling more robust adaptation to novel environments through learned exploration strategies.

Large language model based multi-agent systems (MAS) have unlocked significant advancements in tackling complex problems, but their increasing capability introduces a structural fragility that makes them difficult to debug. A key obstacle to improving their reliability is the severe scarcity of large-scale, diverse datasets for error attribution, as existing resources rely on costly and unscalable manual annotation. To address this bottleneck, we introduce Aegis, a novel framework for Automated error generation and attribution for multi-agent systems. Aegis constructs a large dataset of 9,533 trajectories with annotated faulty agents and error modes, covering diverse MAS architectures and task domains. This is achieved using a LLM-based manipulator that can adaptively inject context-aware errors into successful execution trajectories. Leveraging fine-grained labels and the structured arrangement of positive-negative sample pairs, Aegis supports three different learning paradigms: Supervised Fine-Tuning, Reinforcement Learning, and Contrastive Learning. We develop learning methods for each paradigm. Comprehensive experiments show that trained models consistently achieve substantial improvements in error attribution. Notably, several of our fine-tuned LLMs demonstrate performance competitive with or superior to proprietary models an order of magnitude larger, validating our automated data generation framework as a crucial resource for developing more robust and interpretable multi-agent systems.

This paper presents a unified framework for simulating and visualizing dynamic fracture phenomena in extreme mechanical collisions using multi-view image inputs. While existing methods primarily address elastic deformations at contact surfaces, they fail to capture the complex physics of extreme collisions, often producing non-physical artifacts and material adhesion at fracture interfaces. Our approach integrates two key innovations: (1) an enhanced Collision Material Point Method (Collision-MPM) with momentum-conserving interface forces derived from normalized mass distributions, which effectively eliminates unphysical adhesion in fractured solids; and (2) a fracture-aware 3D Gaussian continuum representation that enables physically plausible rendering without post-processing. The framework operates through three main stages: First, performing implicit reconstruction of collision objects from multi-view images while sampling both surface and internal particles and simultaneously learning surface particle Gaussian properties via splatting; Second, high-fidelity collision resolution using our improved Collision-MPM formulation; Third, dynamic fracture tracking with Gaussian attribute optimization for fracture surfaces rendering. Through comprehensive testing, our framework demonstrates significant improvements over existing methods in handling diverse scenarios, including homogeneous materials, heterogeneous composites, and complex multi-body collisions. The results confirm superior physical accuracy, while maintaining computational efficiency for rendering.

While previous AI Scientist systems can generate novel findings, they often lack the focus to produce scientifically valuable contributions that address pressing human-defined challenges. We introduce DeepScientist, a system designed to overcome this by conducting goal-oriented, fully autonomous scientific discovery over month-long timelines. It formalizes discovery as a Bayesian Optimization problem, using a cumulative Findings Memory to intelligently balance the exploitation of promising avenues with the exploration of novel hypotheses. Consuming over 20,000 GPU hours, the system generated about 5,000 unique ideas and experimentally validated approximately 1100, ultimately surpassing human-designed 2025 state-of-the-art (SOTA) methods on three frontier AI tasks by 183.7\%, 1.9\%, and 7.9\%. Crucially, this was achieved by autonomously redesigning core methodologies, not merely recombining existing techniques. In a striking demonstration, the system achieved progress on AI text detection in just two weeks that is comparable to three years of cumulative human research. This work provides the first large-scale evidence of an AI achieving discoveries that progressively surpass human SOTA on scientific tasks, producing valuable findings that genuinely push the frontier forward.

Process Reward Models (PRMs) have recently emerged as a powerful framework for enhancing the reasoning capabilities of large reasoning models (LRMs), particularly in the context of test-time scaling (TTS). However, their potential for supervising LRMs on tabular reasoning domains remains underexplored. Through detailed empirical analyses, we identify that existing PRMs, though widely adopted for supervising text-only reasoning steps, struggle with table-specific operations such as sub-table retrieval and schema interaction, leading to critical performance bottlenecks. To address this limitation, we propose TaTToo, a novel table-grounded PRM framework that (i) reasons explicitly over tabular reasoning steps and (ii) integrates tool-based verification to provide precise reward supervision. Concretely, we first design a scalable data curation pipeline that constructs over 60k high-quality step-level annotations by integrating table verification rationales with tool-based executions. Building on the collected data, we train TaTToo with a dual-stage paradigm: cold-start supervised fine-tuning to capture tool-use reasoning patterns, followed by reinforcement learning with tool-grounded reward shaping to align our model with table-based verification. We provide a comprehensive evaluation of the policy improvement induced by our newly designed PRM. Across 5 challenging tabular reasoning benchmarks covering numerical reasoning, fact-checking, and data analysis, TaTToo improves downstream policy LRMs by 30.9\% at inference, surpasses strong PRM baselines such as Qwen-2.5-Math-PRM-72B with only 8B parameters, and demonstrates strong generalizability across diverse TTS strategies.

We propose a framework for conditional vector quantile regression (CVQR) that combines neural optimal transport with amortized optimization, and apply it to multivariate conformal prediction. Classical quantile regression does not extend naturally to multivariate responses, while existing approaches often ignore the geometry of joint distributions. Our method parameterizes the conditional vector quantile function as the gradient of a convex potential implemented by an input-convex neural network, ensuring monotonicity and uniform ranks. To reduce the cost of solving high-dimensional variational problems, we introduce amortized optimization of the dual potentials, yielding efficient training and faster inference. We then exploit the induced multivariate ranks for conformal prediction, constructing distribution-free predictive regions with finite-sample validity. Unlike coordinatewise methods, our approach adapts to the geometry of the conditional distribution, producing tighter and more informative regions. Experiments on benchmark datasets show improved coverage–efficiency trade-offs compared to baselines, highlighting the benefits of integrating neural optimal transport with conformal prediction.

Vision-Language Models (VLMs) have enabled computer use agents (CUAs) that operate GUIs autonomously, showing great potential, yet progress is limited by the lack of large-scale, open-source computer use data and foundation models. In this work, we introduce ScaleCUA, a step toward scaling open-source CUAs. It offers a large-scale dataset spanning 6 operating systems and 3 task domains, built via a closed-loop pipeline uniting automated agents with human experts. Trained on this scaled-up data, ScaleCUA can operate seamlessly across platforms. Specifically, it delivers strong gains over baselines (+26.6 on WebArena-Lite-v2, +10.7 on ScreenSpot-Pro) and sets new state-of-the-art results (94.4% on MMBench-GUI L1-Hard, 60.6% on OSWorld-G, 47.4% on WebArena-Lite-v2). These findings underscore the power of data-driven scaling for general-purpose computer use agents. We will release data, models, and code to advance future research.

Representation alignment helps generation by distilling representations from a pretrained vision encoder to intermediate diffusion features. We investigate a fundamental question - `what aspect of the target representation matters for generation, its global information (measured by Imagenet1K accuracy) or its spatial structure (pairwise cosine similarity between patch tokens)''? Prevalent wisdom holds that stronger global performance leads to better generation as a target representation. To study this, we first perform a large-scale empirical analysis across 27 different vision encoders and different model scales. The results are surprising - spatial structure, rather than global performance drives the generation performance of a target representation. To further study this, we introduce two straightforward modifications, which specifically accentuate the transfer of spatial information. We replace the standard MLP projection layer in REPA with a simple convolution layer and introduce a spatial normalization layer for the external representation. Surprisingly, our simple method (implemented in <4 lines of code), termed iREPA, consistently improves convergence speed of REPA, across a diverse set of vision encoders, model sizes, and training variants (such as REPA, REPA-E, meanflow, JiT etc). Our work motivates revisiting the fundamental working mechanism of representational alignment and how it can be leveraged for improved training of generative models.

Writing GPU kernels is a challenging task and critical for AI systems' efficiency. It is also highly iterative: domain experts write code and improve performance through execution feedback. Moreover, it presents verifiable rewards like correctness and speedup, making it a natural environment to apply Reinforcement Learning (RL). To explicitly incorporate the iterative nature of this process into training, we develop a flexible multi-turn RL recipe that addresses unique challenges encountered in real-world settings, such as learning from long trajectories and effective reward attribution across turns. We present Kevin - K(ernel D)evin, the first model trained with multi-turn RL for CUDA kernel generation and optimization. In our evaluation setup, Kevin shows significant gains over its base model (QwQ-32B), improving correctness of generated kernels (in pure CUDA) from 56% to 82% and mean speedup from 0.53x to 1.10x of baseline (PyTorch Eager), and surpassing frontier models like o4-mini (0.78x). Finally, we study its behavior across test-time scaling axes: we found scaling serial refinement more beneficial than parallel sampling. In particular, when given more refinement turns, Kevin shows a higher rate of improvement.

Human beings naturally utilize multiple reasoning modalities to learn and solve logical problems, i.e., different representational formats such as natural language, code, and symbolic logic. In contrast, most existing LLM-based approaches operate with a single reasoning modality during training, typically natural language. Although some methods explored modality selection or augmentation at inference time, the training process remains modality-blind, limiting synergy among modalities. To fill in this gap, we propose Mixture-of-Thought (MoT), a framework that enables LLMs to reason across three complementary modalities: natural language, code, and a newly introduced symbolic modality, truth-table, which systematically enumerates logical cases and partially mitigates key failure modes in natural language reasoning. MoT adopts a two-phase design: (1) self-evolving MoT training, which jointly learns from filtered, self-generated rationales across modalities; and (2) MoT inference, which fully leverages the synergy of three modalities to produce better predictions. Experiments on logical reasoning benchmarks including FOLIO and ProofWriter demonstrate that our MoT framework consistently and significantly outperforms strong LLM baselines with single-modality chain-of-thought approaches, achieving up to +11.7pp average accuracy gain. Further analyses show that our MoT framework benefits both training and inference stages; that it is particularly effective on harder logical reasoning problems; and that different modalities contribute complementary strengths, with truth-table reasoning helping to overcome key bottlenecks in natural language inference.

Understanding human behavior traits is central to applications in human-computer interaction, computational social science, and personalized AI systems. Such understanding often requires integrating multiple modalities to capture nuanced patterns and relationships. However, existing resources rarely provide datasets that combine behavioral descriptors with complementary modalities such as facial attributes and biographical information. To address this gap, we present PersonaX, a curated collection of multimodal datasets designed to enable comprehensive analysis of public traits across modalities. PersonaX consists of (1) CelebPersona, featuring 9444 public figures from diverse occupations, and (2) AthlePersona, covering 4181 professional athletes across 7 major sports leagues. Each dataset includes behavioral trait assessments inferred by three high-performing large language models, alongside facial imagery and structured biographical features. We analyze PersonaX at two complementary levels. First, we abstract high-level trait scores from text descriptions and apply five statistical independence tests to examine their relationships with other modalities. Second, we introduce a novel causal representation learning (CRL) framework tailored to multimodal and multi-measurement data, providing theoretical identifiability guarantees. Experiments on both synthetic and real-world data demonstrate the effectiveness of our approach. By unifying structured and unstructured analysis, PersonaX establishes a foundation for studying LLM-inferred behavioral traits in conjunction with visual and biographical attributes, advancing multimodal trait analysis and causal reasoning. The code is available at https://github.com/lokali/PersonaX.

Large language models (LLMs) have recently demonstrated remarkable progress in formal theorem proving. Yet their ability to serve as practical assistants for mathematicians, filling in missing steps within complex proofs, remains underexplored. We identify this challenge as the task of subgoal completion, where an LLM must discharge short but nontrivial proof obligations left unresolved in a human-provided sketch. To study this problem, we introduce FormalML, a Lean 4 benchmark built from foundational theories of machine learning. Using a translation tactic that converts procedural proofs into declarative form, we extract 4,937 problems spanning optimization and probability inequalities, with varying levels of difficulty. FormalML is the first subgoal completion benchmark to combine premise retrieval and complex research-level contexts. Evaluation of state-of-the-art provers highlights persistent limitations in accuracy and efficiency, underscoring the need for more capable LLM-based theorem provers for effective subgoal completion.

3D Gaussian Splatting (3DGS) has emerged as a powerful representation for high-quality rendering across a wide range of applications. However, its high computational demands and large storage costs pose significant challenges for deployment on mobile devices. In this work, we propose a mobile-tailored real-time Gaussian Splatting method, dubbed Mobile-GS, enabling efficient inference of Gaussian Splatting on edge devices. Specifically, we first identify alpha blending as the primary computational bottleneck, since it relies on the time-consuming Gaussian depth sorting process. To solve this issue, we propose a depth-aware order-independent rendering scheme that eliminates the need for sorting, thereby substantially accelerating rendering. Although this order-independent rendering improves rendering speed, it may introduce transparency artifacts in regions with overlapping geometry due to the scarcity of rendering order. To address this problem, we propose a neural view-dependent enhancement strategy, enabling more accurate modeling of view-dependent effects conditioned on viewing direction, 3D Gaussian geometry, and appearance attributes. In this way, Mobile-GS can achieve both high-quality and real-time rendering. Furthermore, to facilitate deployment on memory-constrained mobile platforms, we propose first-degree spherical harmonics distillation, a neural vector quantization technique, and a contribution-based pruning strategy to reduce the number of Gaussian primitives and compress the 3D Gaussian representation with the assistance of neural networks. Extensive experiments demonstrate that our proposed Mobile-GS achieves real-time rendering and compact model size while preserving high visual quality, making it well-suited for mobile applications.

Evaluation of Text to Speech (TTS) systems is challenging and resource-intensive. Subjective metrics such as Mean Opinion Score (MOS) are not easily comparable between works. Objective metrics are frequently used, but rarely validated against subjective ones. Both kinds of metrics are challenged by recent TTS systems capable of producing synthetic speech indistinguishable from real speech. In this work, we introduce Text to Speech Distribution Score 2 (TTSDS2), a more robust and improved version of TTSDS. Across a range of domains and languages, it is the only one out of 16 compared metrics to correlate with a Spearman correlation above 0.50 for every domain and subjective score evaluated. We also release a range of resources for evaluating synthetic speech close to real speech: A dataset with over 11,000 subjective opinion score ratings; a pipeline for recreating a multilingual test dataset to avoid data leakage; and a benchmark for TTS in 14 languages.

The goal of generative diffusion models is to align the learned distribution with the real data distribution through gradient score matching. However, inherent limitations of current generative models lead to an inevitable gap between generated data and real data. To address this, we propose Weak-to-Strong Diffusion (W2SD), a novel framework that utilizes the estimated gap between existing weak and strong models (i.e., weak-to-strong gap) to bridge the gap between an ideal model and a strong model. By employing a reflective operation that alternates between denoising and inversion with weak-to-strong gap, W2SD steers latent variables along sampling trajectories toward regions of the real data distribution. W2SD is highly flexible and broadly applicable, enabling diverse improvements through the strategic selection of weak-to-strong model pairs (e.g., DreamShaper vs. SD1.5, good experts vs. bad experts in MoE). Extensive experiments demonstrate that W2SD significantly improves human preference, aesthetic quality, and prompt adherence, achieving significantly improved performance across various modalities (e.g., image, video), architectures (e.g., UNet-based, DiT-based, MoE), and benchmarks. For example, Juggernaut-XL with W2SD can improve with the HPSv2 winning rate up to 90\% over the original results. Moreover, the performance gains achieved by W2SD markedly outweigh its additional computational overhead, while the cumulative improvements from different weak-to-strong gap further solidify its practical utility and deployability.

Electrophysiological (ExG) signals offer valuable insights into human physiology, yet building foundation models that generalize across everyday tasks remains challenging due to two key limitations: (i) insufficient data diversity, as most ExG recordings are collected in controlled labs with bulky, expensive devices; and (ii) task-specific model designs that require tailored processing (i.e., targeted frequency filters) and architectures, which limit generalization across tasks. To address these challenges, we introduce an approach for scalable, task-agnostic ExG monitoring in the wild. We collected 50 hours of unobtrusive free-living ExG data with an earphone-based hardware prototype to narrow the data diversity gap. At the core of our approach is Physiology-informed Multi-band Tokenization (PiMT), which decomposes ExG signals into 12 physiology-informed tokens, followed by a reconstruction task to learn robust representations. This enables adaptive feature recognition across the full frequency spectrum while capturing task-relevant information. Experiments on our new DailySense dataset—the first to enable ExG-based analysis across five human senses—together with four public ExG benchmarks, demonstrate that PiMT consistently outperforms state-of-the-art methods across diverse tasks.

Distilling robust reasoning capabilities from large language models (LLMs) into smaller, computationally efficient student models remains an unresolved challenge. Despite recent advances, distilled models frequently suffer from superficial pattern memorization and subpar generalization. To overcome these limitations, we introduce a novel distillation framework that moves beyond simple mimicry to instill a deeper conceptual understanding. Our framework features two key innovations. \underline{\textit{First}}, to address pattern memorization, Explanatory Inversion (EI) generates targeted ``explanatory probes'' that compel the student to articulate the underlying logic behind an answer, rather than just memorizing it. \underline{\textit{Second}}, to improve generalization, Explanatory GRPO (\texttt{EXGRPO}) uses a reinforcement learning algorithm with a novel Dialogue Structure Utility Bonus, which explicitly rewards the student for maintaining a coherent reasoning process across these probes. Extensive evaluations on 12 datasets demonstrate significant improvements. Using Gemma-7b as the student model, our method yields an average \textbf{20.39\%} increase over zero-shot performance and a \textbf{6.02\%} improvement over the state-of-the-art distillation baselines. Moreover, models distilled with our method show remarkable training efficiency (e.g., surpassing vanilla fine-tuning with \textbf{10-25\%} training data) and strong generalization to out-of-distribution tasks.

Denoising diffusions sample from a probability distribution $\mu$ in $\mathbb{R}^d$ by constructing a stochastic process $(\hat{\mathbf{x}}_t:t\ge 0)$ in $\mathbb{R}^d$ such that $\hat{\mathbf{x}}_0$ is easy to sample, but the distribution of $\hat{\mathbf{x}}_T$ at large $T$ approximates $\mu$. The drift $\mathbf{m}:\mathbb{R}^{d}\times\mathbb{R}\to\mathbb{R}^d$ of this diffusion process is learned by minimizing a score-matching objective. Is every probability distribution $\mu$, for which sampling is tractable, also amenable to sampling via diffusions? We address this question by studying its relation to information-computation gaps in statistical estimation. Earlier work in this area constructs broad families of distributions $\mu$ for which sampling is easy, but approximating the drift $\mathbf{m}(\mathbf{y},t)$ is conjectured to be intractable, and provides rigorous evidence for intractability. We prove that this implies a failure of sampling via diffusions. First, there exist drifts whose score matching objective is superpolynomially close to the optimum value (among polynomial time drifts) and yet yield samples with distribution that is very far from the target one. Second, any polynomial-time drift that is also Lipschitz continuous results in equally incorrect sampling. We instantiate our results on the toy problem of sampling a sparse low-rank matrix, and further demonstrate empirically the failure of diffusion-based sampling. Our work implies that caution should be used in adopting diffusion sampling when other approaches are available.

Figuring out how neural language models comprehend syntax acts as a key to revealing how they understand languages. We systematically analyzed methods for finding syntax structures in models, namely probing, and found limitations yet widely exist in previous probing practice. We proposed a method capable of estimating mutual information (MI) and extracting dependency trees from attention scores in a mathematical-rigorous way, requiring no additional network training effort. Compared with previous approaches, it has a much simpler model, while being able to probe more complex dependency trees, also transparent for fine-grained explanation. We tested our method on several open-source LLMs and demonstrated its effectiveness by systematically comparing it with a great many competitive baselines. Several informative conclusions can be drawn by further analysis of the results, shedding light on our method’s explanatory potential. Our code is released at https://github.com/ChristLBUPT/IPBP.

DINOv2 sees the world well enough to guide robots and segment images, but we still do not know what it sees. We conduct the first comprehensive analysis of DINOv2’s representational structure using overcomplete dictionary learning, extracting over 32,000 visual concepts in what constitutes the largest interpretability demonstration for any vision foundation model to date. This method provides the backbone of our study, which unfolds in three parts. In the first part, we analyze how different downstream tasks recruit concepts from our learned dictionary, revealing functional specialization: classification exploits “Elsewhere” concepts that fire everywhere except on target objects, implementing learned negations; segmentation relies exclusively on boundary detectors forming coherent subspaces; depth estimation draws on three distinct monocular cue families matching visual neuroscience principles. Turning to concept geometry and statistics, we find the learned dictionary deviates from ideal near-orthogonal (Grassmannian) structure, exhibiting higher coherence than random baselines. Concept atoms are not aligned with the neuron basis, confirming distributed encoding. We discover antipodal concept pairs that encode opposite semantics (e.g., “white shirt” vs “black shirt”), creating signed semantic axes. Separately, we identify concepts that activate exclusively on register tokens, revealing these encode global scene properties like motion blur and illumination. Across layers, positional information collapses toward a 2D sheet, yet within single images token geometry remains smooth and clustered even after position is removed, putting into question a purely sparse-coding view of representation. To resolve this paradox, we advance a different view: tokens are formed by combining convex mixtures of a few archetypes (e.g., a rabbit among animals, brown among colors, fluffy among textures). Multi-head attention directly implements this construction, with activations behaving like sums of convex regions. In this picture, concepts are expressed by proximity to landmarks and by regions—not by unbounded linear directions. We call this the Minkowski Representation Hypothesis (MRH), and we examine its empirical signals and consequences for how we study, steer, and interpret vision-transformer representations.

While Language Models (LMs) have made significant progress in automating machine learning engineering (MLE), the acquisition of high-quality MLE training data is significantly constrained. Current MLE benchmarks suffer from low scalability and limited applicability because they rely on static, manually curated tasks that demand extensive time and manual effort to produce. We introduce MLE-Smith, a fully automated multi-agent pipeline, to transform raw datasets into competition-style MLE challenges through an efficient generate--verify--execute paradigm for scaling MLE tasks with verifiable quality, real-world usability and rich diversity. The proposed multi-agent pipeline in MLE-Smith drives structured task design and standardized refactoring, coupled with a hybrid verification mechanism that enforces strict structural rules and high-level semantic soundness. It further validates empirical solvability and real-world fidelity through interactive execution. We apply MLE-Smith to 224 of real-world datasets and generates 606 tasks spanning multiple categories, objectives, and modalities, demonstrating that MLE-Smith can work effectively across a wide range of real-world datasets. Evaluation on generated tasks shows that the performance of eight mainstream and cutting-edge LLMs on MLE-Smith tasks is strongly correlated with their performance on carefully human-designed tasks, highlighting the effectiveness of the MLE-Smith in scaling up MLE tasks while maintaining task quality.

Instruction tuning improves the LLMs performance but depends on high-quality training data. Recently, LLMs have been used to synthesize data, enhancing training with seeds like question-answer (QA) pairs. However, this synthesis often results in instruction examples similar to the seeds, lacking diversity and biasing real applications. Thus, we propose to extract instruction tuning data from web corpus with much rich knowledge. The most straightforward strategy is to quickly retrieve domain specific documents from the corpus and then extract all QA pairs of these documents for tuning LLMs, which has two main limitations. (1) Extracting all QA pairs using LLMs is prohibitively expensive; and (2) These extracted pairs are not all beneficial for the downstream applications, and incorporating all of them for tuning may even hurt the model performance. To overcome the limitations, we introduce $\texttt{EQUAL}$, an $\textbf{E}$ffective and scalable data extraction framework that iteratively interleaves document selection and extract high-$\textbf{QUAL}$ity QA pairs to optimize instruction tuning. $\texttt{EQUAL}$ first clusters the document set based on the embeddings generated by contrastive learning. Then it leverages the multi-armed bandit based strategy to quickly identify document clusters where can extract high-quality QA pairs for training. This iterative framework significantly reduces computational costs while improving model performance much. Experiments on AutoMathText, KnowledgePile and StackOverflow across 13 downstream tasks demonstrate that $\texttt{EQUAL}$ reduces computational costs by 5–10$\times$ while improving accuracy by 2.5\% on LLaMA-3.1-8B, Qwen2.5-7B and Mistral-7B. Code and data is available at https://anonymous.4open.science/r/EQUAL-DD20.

Video DiTs have advanced video generation, yet they still struggle to model multi-instance or subject-object interactions. This raises a key question: How do these models internally represent interactions? To answer this, we curate MATRIX-11K, a video dataset with interaction-aware captions and multi-instance mask tracks. Using this dataset, we conduct a systematic analysis that formalizes two perspectives of video DiTs: semantic grounding, via video-to-text attention, which evaluates whether noun and verb tokens capture instances and their relations; and semantic propagation, via video-to-video attention, which assesses whether instance bindings persist across frames. We find both effects concentrate in a small subset of interaction-dominant layers. Motivated by this, we introduce MATRIX, a simple and effective regularization that aligns attention in specific layers of video DiTs with multi-instance mask tracks from the MATRIX-11K dataset, enhancing both grounding and propagation. We further propose InterGenEval, an evaluation protocol for interaction-aware video generation. In experiments, MATRIX improves both interaction fidelity and semantic alignment while reducing drift and hallucination. Extensive ablations validate our design choices. Codes and weights will be released.

Reinforcement learning (RL) is a framework for solving sequential decision-making problems. In this work, we demonstrate that, surprisingly, RL emerges during the inference time of large language models (LLMs), a phenomenon we term in-context RL (ICRL). To reveal this capability, we introduce a simple multi-round prompting framework, we call ICRL prompting, for inference-time self-improvement. The goal of ICRL prompting is to guide LLMs to perform reinforcement learning during inference for self-improvement on a given task. After each response, the model receives numerical scalar feedback, denoted as a reward. In the next round, we prompt the LLM again together with a context that concatenates all prior responses and their associated rewards. We consistently observe that response quality improves as the context grows. In other words, the LLM can optimize scalar reward signals during inference, exhibiting behavior analogous to reinforcement learning. We evaluate ICRL prompting on Game of 24, creative writing, ScienceWorld, and Olympiad-level math competitions (AIME and HMMT), demonstrating significant improvements over baselines such as Self-Refine and Reflexion. Notably, even when the reward signals are generated by the same LLM, ICRL prompting still improves performance, highlighting a promising new paradigm for test-time scaling.

Tree-Sliced methods have emerged as an efficient and expressive alternative to the traditional Sliced Wasserstein distance, replacing one-dimensional projections with tree-structured metric spaces and leveraging a splitting mechanism to better capture the underlying topological structure of integration domains while maintaining low computational cost. At the core of this framework is the Tree-Sliced Wasserstein (TSW) distance, defined over probability measures in Euclidean spaces, along with several variants designed to enhance its performance. A fundamental distinction between SW and TSW lies in their sampling strategies—a component explored in the context of SW but often overlooked in comparisons. This omission is significant: whereas SW relies exclusively on directional projections, TSW incorporates both directional and positional information through its tree-based construction. This enhanced spatial sensitivity enables TSW to reflect the geometric structure of the underlying data more accurately. Building on this insight, we propose a novel variant of TSW that explicitly leverages positional information in its design. Inspired by the classical Fermat–Weber problem—which seeks a point minimizing the sum of distances to a given set of points—we introduce the Fermat–Weber Tree-Sliced Wasserstein (FW-TSW) distance. By incorporating geometric median principles into the tree construction process, FW-TSW notably further improves the performance of TSW while preserving its low computational cost. These improvements are empirically validated across diverse experiments, including diffusion model training and gradient flow. Our code is available at https://github.com/thanhquangtran/FW-TSW.

Reinforcement learning with verifiable rewards (RLVR) has emerged as a promising paradigm for enhancing the reasoning capabilities of large language models (LLMs). In this context, models explore reasoning trajectories and exploit rollouts with correct answers as positive signals for policy optimization. However, these rollouts might involve flawed patterns such as answer-guessing and jump-in-reasoning. Such flawed-positive rollouts are rewarded identically to fully correct ones, causing policy models to internalize these unreliable reasoning patterns. In this work, we first conduct a systematic study of flawed-positive rollouts in RL and find that they enable rapid capability gains during the early optimization stage, while constraining reasoning capability later by reinforcing unreliable patterns. Building on these insights, we propose Flawed-Aware Policy Optimization (FAPO), which presents a parameter-free reward penalty for flawed-positive rollouts, enabling the policy to leverage them as useful shortcuts in the warm-up stage, securing stable early gains, while gradually shifting optimization toward reliable reasoning in the later refinement stage. To accurately and comprehensively detect flawed-positive rollouts, we introduce a generative reward model (GenRM) with a process-level reward that precisely localizes reasoning errors. Experiments show that FAPO is effective in broad domains, improving outcome correctness, process reliability, and training stability without increasing the token budget.

The exponential growth in demand for GPU computing resources has created an urgent need for automated CUDA optimization strategies. While recent advances in LLMs show promise for code generation, current state-of-the-art models achieve low success rates in improving CUDA speed. In this paper, we introduce CUDA-L1, an automated reinforcement learning (RL) framework for CUDA optimization that employs a novel contrastive RL algorithm. CUDA-L1 achieves significant performance improvements on the CUDA optimization task: trained on NVIDIA A100, it delivers an average speedup of {\bf ×3.12} with a median speedup of {\bf ×1.42} against default baselines over across all 250 CUDA kernels of KernelBench, with peak speedups reaching {\bf ×120}. In addition to the default baseline provided by KernelBench, CUDA-L1 demonstrates {\bf ×2.77} over Torch Compile, {\bf ×2.88} over Torch Compile with reduce overhead, and {\bf ×2.81} over CUDA Graph implementations. Furthermore, the model also demonstrates portability across GPU architectures, achieving average speedups of {\bf ×3.85} (median {\bf ×1.32}) on H100, {\bf ×3.13} (median {\bf ×1.31}) on L40, {\bf ×2.51} (median {\bf ×1.18}) on RTX 3090, and {\bf ×2.38} (median {\bf ×1.34}) on H20 despite being optimized specifically for A100. Beyond these benchmark results, CUDA-L1 demonstrates several properties: CUDA-L1 1) discovers a variety of CUDA optimization techniques and learns to combine them strategically to achieve optimal performance; 2) uncovers fundamental principles of CUDA optimization, such as the multiplicative nature of optimizations; 3) identifies non-obvious performance bottlenecks and rejects seemingly beneficial optimizations that actually harm performance. The capabilities demonstrate that, RL can transform an initially poor-performing LLM into an effective CUDA optimizer through speedup-based reward signals alone, without human expertise or domain knowledge. In this process, it identifies CUDA optimization patterns, discovers new techniques, synthesizes them to achieve speedups, and more importantly, extends the acquired reasoning abilities to new kernels. This paradigm opens possibilities for automated optimization of CUDA operations, and holds promise to substantially promote GPU efficiency and alleviate the rising pressure on GPU computing resources.

We propose Segment Concept (SeC), a concept-driven video object segmentation (VOS) framework that shifts from conventional feature matching to the progressive construction and utilization of high-level, object-centric representations. SeC employs Large Vision-Language Models (LVLMs) to integrate visual cues across diverse frames, constructing robust conceptual priors. To balance semantic reasoning with computational overhead, SeC forwards the LVLMs only when a new scene appears, injecting concept-level features at those points. To rigorously assess VOS methods in scenarios demanding high-level conceptual reasoning and robust semantic understanding, we introduce the Semantic Complex Scenarios Video Object Segmentation benchmark (SeCVOS). SeCVOS comprises 160 manually annotated multi-scenario videos designed to challenge models with substantial appearance variations and dynamic scene transformations. Empirical evaluations demonstrate that SeC substantially outperforms state-of-the-art approaches, including SAM 2 and its advanced variants, on both SeCVOS and standard VOS benchmarks. In particular, SeC achieves an 11.8-point improvement over SAM 2.1 on SeCVOS, establishing a new state-of-the-art in concept-aware VOS.

Discrete diffusion models have demonstrated great promise in modeling various sequence data, ranging from human language to biological sequences. Inspired by the success of RL in language models, there is growing interest in further improving the models by alignment with a certain reward. In this work, we propose an offline preference optimization method to approach trajectory alignment for discrete diffusion models. Instead of applying the reward on the final output and backpropagating the gradient to the entire denoising process, we decompose the problem into a set of stepwise alignment objectives by matching the per-step posterior. This framework enables efficient diffusion optimization, is compatible with arbitrary reward functions, and importantly, yields an equivalent optimal solution under additive factorization of the trajectory reward. Experiments across multiple domains including DNA sequence design, protein inverse folding, and language modeling consistently demonstrate the superiority of our approach. Notably, it achieves an up to 12\% improvement over the most competitive RL-based baseline in terms of predicted activity on DNA sequence design, and further improves the GSM8K score from 78.6 to 81.2 on LLaDA-8B-Instruct for language modeling.

Large language models (LLMs) have achieved impressive performance on knowledge-intensive tasks, yet they often struggle with multi-step reasoning due to the unstructured nature of retrieved context. While retrieval-augmented generation (RAG) methods provide external information, the lack of explicit organization among retrieved passages limits their effectiveness, leading to brittle reasoning pathways. Recent interpretability studies highlighting the importance of structured intermediate reasoning further align with this perspective. We propose Retrieval-And-Structuring (RAS), a framework that dynamically constructs question-specific knowledge graphs through iterative retrieval and structured knowledge building. RAS interleaves targeted retrieval planning with incremental graph construction, enabling models to assemble and reason over evolving knowledge structures tailored to each query. On seven knowledge-intensive benchmarks, RAS consistently outperforms strong baselines, achieving up to 8.7\% and 7.0\% gains with proprietary and open-source LLMs, respectively. Our results demonstrate that dynamic, question-specific knowledge structuring offers a robust path to improving reasoning accuracy and robustness in language model generation.

Multi-Agent System (MAS) and Reinforcement Learning (RL) are both widely adopted to improve large language model (LLM) agentic performance. MAS strengthens task-specialized performance via role-based orchestration; RL leverages environment rewards to train stronger policies, such as Group Relative Policy Optimization (GRPO)-style optimization. Yet applying on-policy RL training to MAS is underexplored. While promising, it poses several challenges. On the algorithm side, Standard GRPO grouping assumptions fail in MAS because prompts differ by role and turn. On the system side, the training system needs to support MAS-workflow-based rollouts and on-policy updates for both single and multiple policy models. To address these issues, we introduce AT-GRPO, consisting of (i) an Agent- and Turn-wise grouped RL algorithm tailored for MAS and (ii) a system to support both single-policy and multi-policy training. Across game, plan, coding, and math tasks, AT-GRPO demonstrates substantial performance gains across diverse domains. Especially on long-horizon planning tasks, AT-GRPO boosts accuracy from a 14.0–47.0% single-agent RL baseline to 96.0–99.5%. Furthermore, it improves reasoning performance, with an average gain of 3.87–7.62% on coding and 9.0-17.93% on math. The code are available at https://github.com/pettingllms-ai/PettingLLMs.

Mixed-curvature spaces have emerged as a powerful alternative to their Euclidean counterpart, enabling data representations better aligned with the intrinsic structure of complex datasets. However, comparing probability distributions in such spaces remains underdeveloped: existing measures such as KL divergence and Wasserstein either rely on strong assumptions on distributions or incur high computational costs. The Sliced-Wasserstein (SW) framework provides an alternative approach for constructing distributional distances; however, its reliance on one-dimensional projections limits its ability to capture the geometry of the ambient space. To address this limitation, the Tree-Sliced Wasserstein (TSW) framework employs tree structures as a richer projected space. Motivated by the intuition that such a space is particularly suitable for representing the geometric properties of mixed-curvature manifolds, we introduce the Mixed-Curvature Tree-Sliced Wasserstein (MC-TSW), a novel discrepancy measure that is computationally efficient while faithfully capturing both the topological and geometric structures of mixed-curvature spaces. Specifically, we introduce an adaptation of tree systems and Radon transform to mixed-curvature spaces, which yields a closed-form solution for the optimal transport problem on the tree system. We further provide theoretical analysis on the properties of the Radon transform and the MC-TSW distance. Experimental results demonstrate that MC-TSW improves distributional comparisons over product-space-based distance and line-based baselines, and that mixed-curvature representations consistently outperform constant-curvature alternatives, highlighting their importance for modeling complex datasets.

Modeling human–human interactions from text remains challenging because it requires not only realistic individual dynamics but also precise, text-consistent spatiotemporal coupling between agents. Currently, progress is hindered by 1) limited two-person training data, inadequate to capture the diverse intricacies of two-person interactions; and 2) insufficiently fine-grained text-to-interaction modeling, where language conditioning collapses rich, structured prompts into a single sentence embedding. To address these limitations, we propose our Text2Interact framework, designed to generate realistic, text-aligned human–human interactions through a scalable high-fidelity interaction data synthesizer and an effective spatiotemporal coordination pipeline. First, we present InterCompose, a scalable synthesis-by-composition pipeline that aligns LLM-generated interaction descriptions with strong single-person motion priors. Given a prompt and a motion for an agent, InterCompose retrieves candidate single-person motions, trains a conditional reaction generator for another agent, and uses a neural motion evaluator to filter weak or misaligned samples—expanding interaction coverage without extra capture. Second, we propose InterActor, a text-to-interaction model with word-level conditioning that preserves token-level cues (initiation, response, contact ordering) and an adaptive interaction loss that emphasizes contextually relevant inter-person joint pairs, improving coupling and physical plausibility for fine-grained interaction modeling. Extensive experiments show consistent gains in motion diversity, fidelity, and generalization, including out-of-distribution scenarios and user studies. Code will be released at github.com/Qingxuan-Wu/Text2Interact.

Kolmogorov-Arnold Networks (KANs) have emerged as a promising alternative to traditional neural networks, offering enhanced interpretability based on the Kolmogorov-Arnold representation theorem. While their empirical success is growing, a theoretical understanding of their training dynamics remains nascent. This paper investigates the optimization of a two-layer KAN in the overparameterized regime, focusing on a simplified yet insightful setting where only the first-layer coefficients are trained via gradient descent. Our main result establishes that, provided the network is sufficiently wide, this training method is guaranteed to converge to a global minimum and achieve zero training error. Furthermore, we derive a novel, fine-grained convergence rate that explicitly connects the optimization speed to the structure of the data labels through the eigenspectrum of the KAN Tangent Kernel (KAN-TK). Our analysis reveals a key advantage of this architecture: guaranteed convergence is achieved with a hidden layer width of $m=\mathcal{O}(n^2)$, a significant polynomial improvement over the $m=\mathcal{O}(n^6)$ requirement for classic two-layer neural networks using ReLU activation functions and analyzed within the same Tangent Kernel framework. We validate our theoretical findings with numerical experiments that corroborate our predictions on convergence speed and the impact of label structure.

Spatial reasoning, the ability to ground language in 3D understanding, remains a persistent challenge for Vision-Language Models (VLMs). We identify two fundamental bottlenecks: \textit{inadequate} 3D understanding capabilities stemming from 2D-centric pre-training, and reasoning failures induced by \textit{redundant} 3D information. To address these, we first construct a Minimal Sufficient Set (MSS) of information before answering a given question: a \textit{compact} selection of 3D perception results from \textit{expert models}. We introduce \textbf{MSSR} (Minimal Sufficient Spatial Reasoner), a dual-agent framework that implements this principle. A \textit{Perception Agent} programmatically queries 3D scenes using a versatile perception toolbox to extract sufficient information, including a novel \textbf{SOG} (Situated Orientation Grounding) module that robustly extracts language-grounded directions. A \textit{Reasoning Agent} then iteratively refines this information to pursue minimality, pruning redundant details and requesting missing ones in a closed loop until the MSS is curated. Extensive experiments demonstrate that our method, by explicitly pursuing both sufficiency and minimality, significantly improves accuracy and achieves state-of-the-art performance across two challenging benchmarks. Furthermore, our framework produces interpretable reasoning paths, offering a promising source of high-quality training data for future models. Source code will be made publicly available.

Recent advancements in LLM-based information-seeking agents have achieved record-breaking performance on established benchmarks. However, these agents remain heavily reliant on search-engine-indexed knowledge, leaving a critical blind spot: Unindexed Information Seeking (UIS). This paper identifies and explores the UIS problem, where vital information is not captured by search engine crawlers, such as overlooked content, dynamic webpages, and embedded files. Despite its significance, UIS remains an underexplored challenge. To address this gap, we introduce UIS-QA, the first dedicated UIS benchmark, comprising 110 expert-annotated QA pairs. Notably, even state-of-the-art agents experience a drastic performance drop on UIS-QA (e.g., from 70.90 on GAIA and 46.70 on BrowseComp-zh to 24.55 on UIS-QA), underscoring the severity of the problem. To mitigate this, we propose UIS-Digger, a novel multi-agent framework that incorporates dual-mode browsing and enables simultaneous webpage searching and file parsing. With a relatively small $\sim$30B-parameter backbone LLM optimized using SFT and RFT training strategies, UIS-Digger sets a strong baseline at 26.36\%, outperforming systems integrating sophisticated LLMs such as O3 and GPT-4.1. This demonstrates the importance of proactive interaction with unindexed sources for effective and comprehensive information-seeking. Our work not only uncovers a fundamental limitation in current agent evaluation paradigms but also provides the first toolkit for advancing UIS research, defining a new and promising direction for robust information-seeking systems.

We study the problem of training and fine-tuning expressive policies with online reinforcement learning (RL) given an offline dataset. Training expressive policy classes with online RL present a unique challenge of stable value maximization. Unlike simpler Gaussian policies commonly used in online RL, expressive policies like diffusion and flow-matching policies are parameterized by a long denoising chain, which hinders stable gradient propagation from actions to policy parameters when optimizing against some value function. Our key insight is that we can address stable value maximization by avoiding direct optimization over value with the expressive policy and instead construct an on-the-fly RL policy to maximize Q-value. We propose Expressive Policy Optimization (EXPO), a sample-efficient online RL algorithm that utilizes an on-the-fly policy to maximize value with two parameterized policies -- a larger expressive base policy trained with a stable imitation learning objective and a light-weight Gaussian edit policy that edits the actions sampled from the base policy toward a higher value distribution. The on-the-fly policy optimizes the actions from the base policy with the learned edit policy and chooses the value maximizing action from the base and edited actions for both sampling and temporal-difference (TD) backup. Our approach yields up to 2-3x improvement in sample efficiency on average over prior methods both in the setting of fine-tuning a pretrained policy given offline data and in leveraging offline data to train online.

Visual generation models have made remarkable progress in creating realistic images from text prompts, yet struggle with complex prompts that specify multiple objects with precise spatial relationships and attributes. Effective handling of such prompts requires explicit reasoning about the semantic content and spatial layout. We present GoT-R1, a framework that applies reinforcement learning to enhance semantic-spatial reasoning in autoregressive visual generation models. Leveraging the natural affinity between autoregressive architectures and sequential reasoning, our approach builds upon the Generation Chain-of-Thought framework to enable models to autonomously discover effective reasoning strategies beyond predefined templates. To achieve this, we propose a dual-stage multi-dimensional reward framework that leverages MLLMs to evaluate both the reasoning process and final output, enabling effective supervision across the entire generation pipeline. The reward system assesses semantic alignment, spatial accuracy, and visual quality in a unified approach. Experimental results demonstrate significant improvements on T2I-CompBench and GenEval benchmark, particularly in compositional tasks involving precise spatial relationships and attribute binding. GoT-R1 advances the state-of-the-art in autoregressive image generation by successfully transferring sophisticated reasoning capabilities from language models to the visual generation domain. Code is available at https://github.com/gogoduan/GoT-R1.

Multimodal Large Language Models (MLLMs) suffer from hallucinations. Existing hallucination evaluation benchmarks are often limited by over-simplified tasks leading to saturated metrics, or insufficient diversity that fails to adequately assess the hallucination extent in state-of-the-art multimodal models. To address this gap, we propose FREAK, a comprehensive multimodal benchmark designed for fine-grained hallucination assessment in MLLMs. Through high-quality photorealistic images featuring fine-grained counter-commonsense edits, FREAK innovatively evaluates hallucination phenomena in detailed visual perception of MLLMs. Extensive experiments on FREAK show severe hallucination issues in SOTA models regarding detailed visual perception. To enable deeper investigation, we curate a controlled subset to indirectly evaluate the model’s ability to perceive target detailed information. Through systematic evaluation of prevailing Chain-of-Thought (CoT) prompting techniques within this task, we reveal critical insights regarding hallucination patterns and model reasoning processes.

In-context learning (ICL) has demonstrated significant performance improvements in transformer-based large models. This study identifies two key factors influencing ICL generalisation under complex non-i.i.d. scenario: algorithmic stability and distributional discrepancy. First, we establish a stability bound for transformer-based models trained with mini-batch gradient descent, revealing how specific optimization configurations interact with the smoothness of the loss landscape to ensure the stability of non-linear Transformers. Next, we introduce a distribution-level discrepancy measure that highlights the importance of aligning the ICL prompt distribution with the training data distribution to achieve effective generalisation. Building on these insights, we derive a generalisation error bound for ICL with asymptotic convergence guarantees, which further reveals that token-wise prediction errors accumulate over time and even lead to generalisation collapse if the prediction length is not properly constrained. Finally, empirical evaluations are provided to validate our theoretical findings.

High-quality visual representation is a long-standing pursuit in computer vision. In the context of multimodal LLMs (MLLMs), feeding higher-resolution images can produce more fine-grained visual tokens. However, it introduces additional computational and design complexity, due to multiple forward passes and post-processing of increased tokens. Before simply adopting a higher resolution, have we truly unlocked the model's full perception capability at a standard resolution? Therefore, we study an interesting problem: how to achieve fine visual perception under lower cost without larger images. We present SigLIP-HD in this work. The core is a highly simple fine-to-coarse supervision design. We enforce the coarse feature of a mid-resolution image to mimic the fine-grained feature of its high-resolution version. We build this framework on the advanced SigLIP 2 model. Our final model produces better visual tokens at exactly the same inference budget. It is validated on extensive MLLM benchmarks and consistently delivers stronger results than our baseline model, especially on OCR-related tasks.

Real-time image segmentation is a key enabler for AR/VR, robotics, drones, and autonomous systems, where tight accuracy, latency, and energy budgets must be met on resource‑constrained edge devices. While U‑Net offers a favorable balance of accuracy and efficiency compared to large transformer‑based models, achieving real‑time performance on high‑resolution input remains challenging due to compute, memory, and power limits. Extreme quantization, particularly binary networks, is appealing for its hardware‑friendly operations. However, two obstacles limit practicality: (1) severe accuracy degradation, and (2) a lack of end‑to‑end implementations that deliver efficiency on general‑purpose GPUs. We make two empirical observations. (1) An explicit zero state is essential: training with zero masking to binary U‑Net weights yields noticeable sparsity. (2) Quantization sensitivity is relatively uniform across layers. Motivated by these findings, we introduce Masked Binary U‑Net (MBU‑Net), obtained through a cost‑aware masking strategy that prioritizes masking where it yields the highest accuracy‑per‑cost, reconciling accuracy with near‑binary efficiency. To realize these gains in practice, we develop a GPU execution framework that maps MBU‑Net to Tensor Cores via a subtractive bit‑encoding scheme, efficiently implementing masked binary weights with binary activations. This design leverages native binary Tensor Core BMMA instructions, enabling high throughput and energy savings on widely available GPUs. Across 3 segmentation benchmarks, MBU‑Net attains near full‑precision accuracy (3\% average drop) while delivering 2.04$\times$ speedup and 3.54$\times$ energy reductions over a 16-bit floating point U‑Net. The code is available at https://github.com/ChunshuWu/MBU-Net.

Exploration remains the key bottleneck for large language model agents trained with reinforcement learning. While prior methods exploit pretrained knowledge, they fail in environments requiring the discovery of novel states. We propose EMPO$^2$, a hybrid RL framework that leverages memory for exploration and combines on- and off-policy updates to make LLMs perform well with memory while also ensuring robustness without it. On ScienceWorld and WebShop, EMPO$^2$ achieves 128.6% and 11.3% improvements over GRPO, respectively. Moreover, in out-of-distribution tests, EMPO$^2$ demonstrates superior adaptability to new tasks, requiring only a few trials with memory and no parameter updates. These results highlight EMPO$^2$ as a promising framework for building more exploratory and generalizable LLM-based agents.

Generative model ecosystems increasingly operate as competitive multi-platform markets, where platforms strategically select models from a shared pool and users with heterogeneous preferences choose among them. Understanding how platforms interact, when market equilibria exist, how outcomes are shaped by model-provider, platforms, and user behavior, and how social welfare is affected is critical for fostering beneficial market environment. In this paper, we formalize a three-layer model-platfrom-user market game and identify conditions for the existence of pure Nash equilibrium. Our analysis shows that market structure, whether platforms converge on similar models or differentiate by selecting distinct ones, depends not only on models’ global average performance but also on their localized attraction to user groups. We further examine welfare outcomes and show that expanding the model pool does not necessarily increase user welfare or market diversity. Finally, we design and evaluate best-response training schemes that allow model-provider to strategically introduce new models into competitive markets.

A molecule’s properties are fundamentally determined by its composition and structure encoded in its molecular graph. Thus, reasoning about molecular properties requires the ability to parse and understand the molecular graph. Large Language Models (LLMs) are increasingly applied to chemistry, tackling tasks such as molecular name conversion, captioning, text-guided generation, and property or reaction prediction. Most existing benchmarks emphasize general chemical knowledge, rely on literature or surrogate labels that risk leakage or bias, or reduce evaluation to multiple-choice questions. We introduce MolecularIQ, a molecular structure reasoning benchmark focused exclusively on symbolically verifiable tasks. MolecularIQ enables fine-grained evaluation of reasoning over molecular graphs and reveals capability patterns that localize model failures to specific tasks and molecular structures. This provides actionable insights into the strengths and limitations of current chemistry LLMs and guides the development of models that reason faithfully over molecular structure.

With the advent of DeepSeek-R1, a new wave of reinforcement learning (RL) methods has emerged that seem to unlock stronger mathematical reasoning. However, a closer look at the open-source ecosystem reveals a critical limitation: with sufficiently many draws (e.g., $\texttt{pass@1024}$), existing base models already solve nearly all questions on widely used math benchmarks such as MATH-500 and AIME 2024. This suggests that the RL fine-tuning methods prevalent in the LLM reasoning literature largely sharpen existing solution modes rather than discovering entirely new ones. Such sharpening stands in contrast to the broader promise of RL: to foster exploration and to acquire new skills. To move beyond this plateau, we introduce MATH-Beyond (MATH-B), a benchmark deliberately constructed to defeat common open-source models of up to 8B parameters even under large sampling budgets. Improving performance on our benchmark via RL requires methods that learn to reason in ways that go beyond base model capabilities in repeated sampling. Since the problems are drawn from subsets of DAPO-Math-17K and DeepScaleR datasets, they remain topically equivalent to standard high-school math. Validating our premise, RL fine-tuned models such as Nemotron-Research-Reasoning-Qwen-1.5B and DeepScaleR-1.5B-Preview perform poorly on MATH-B at $\texttt{pass@1024}$, showing how existing approaches fall short on tackling harder instances. We hope MATH-B will catalyze exploration-driven RL approaches that elicit deeper reasoning capabilities.

AI-assisted programming is rapidly reshaping software development, with large language models (LLMs) enabling new paradigms such as vibe coding and agentic coding. While prior works have focused on prompt design and code generation quality, the broader impact of LLM-driven development on the iterative dynamics of software engineering remains underexplored. In this paper, we conduct large-scale experiments on thousands of algorithmic programming tasks and hundreds of framework selection tasks to systematically investigate how AI-assisted programming interacts with the software ecosystem. Our analysis quantifies a substantial performance asymmetry: mainstream languages and frameworks achieve significantly higher success rates than niche ones. This disparity suggests a feedback loop consistent with the Matthew Effect, where data-rich ecosystems gain superior AI support. While not the sole driver of adoption, current models introduce a non-negligible productivity friction for niche technologies, representing a hidden bias in software evolution.

Current zero-shot unsupervised segmentation methods based on normalized cuts (NCut) face three key limitations. First, they rely on recursive bipartitions with repeated eigen-decompositions, making them prohibitively expensive at scale. Second, each split requires spectral relaxation followed by rounding, introducing layers of approximation where the final partition may diverge from the true NCut objective. Third, recursive bipartitioning offers no principled assurance of producing a stable $K$-way segmentation, and existing heuristics lack convergence guarantees. We propose \textbf{Falcon}, a proximal-gradient solver that directly optimizes the discrete $K$-way NCut objective without spectral relaxation. We prove linear convergence under the \textit{Kurdyka--\L{}ojasiewicz} (KL) property. Falcon computes closed-form gradient scores weighted by cluster volumes and performs row-wise one-hot proximal updates stabilized by inertia. A monotone backtracking scheme adaptively tunes the proximal parameter, ensuring non-decreasing NCut values. This design preserves discrete feasibility, removes repeated eigen-decomposition, and guarantees convergence. Across six benchmarks, Falcon outperforms the strongest official baseline (DiffCut) by wide margins, e.g., +13.2 mIoU on VOC, +27.7 on COCO-Object, and +3.1 on Cityscapes, while remaining competitive on Pascal Context. It also runs up to an order of magnitude faster than recursive NCut and scales more favorably in memory at high resolution, making it practical for larger token grids. By pairing pretrained foundation models with a principled NCut solver, Falcon sets a new state of the art across six benchmarks and achieves the best performance on 17 of 18 benchmark--encoder pairs, underscoring both its robustness and its generality in bridging the gap between unsupervised and supervised segmentation.

Large language model (LLM) simulations of human behavior have the potential to revolutionize the social and behavioral sciences, if and only if they faithfully reflect real human behaviors. Current evaluations of simulation fidelity are fragmented, based on bespoke tasks and metrics, creating a patchwork of incomparable results. To address this, we introduce SimBench, the first large-scale, standardized benchmark for a robust, reproducible science of LLM simulation. By unifying 20 diverse datasets covering tasks from moral decision-making to economic choice across a large global participant pool, SimBench provides the necessary foundation to ask fundamental questions about when, how, and why LLM simulations succeed or fail. We show that the best LLMs today achieve meaningful but modest simulation fidelity (score: 40.80/100), with performance scaling log-linearly with model size but not with increased inference-time compute. We discover an alignment-simulation tradeoff: instruction tuning improves performance on low-entropy (consensus) questions but degrades it on high-entropy (diverse) ones. Models particularly struggle when simulating specific demographic groups. Finally, we demonstrate that simulation ability correlates most strongly with knowledge-intensive reasoning (MMLU-Pro, $r=0.939$). By making progress measurable, we aim to accelerate the development of more faithful LLM simulators.

As large language models (LLMs) scale to billions of parameters, full-parameter fine-tuning becomes compute- and memory-prohibitive. Parameter-efficient fine-tuning (PEFT) mitigates this issue by updating only a small set of task-specific parameters while keeping the base model frozen. Among PEFT approaches, low-rank adaptation (LoRA) is widely adopted; however, it enforces a uniform rank across layers despite substantial variation in layer importance, motivating layerwise rank allocation. Recent adaptive-rank variants (e.g., AdaLoRA) allocate ranks based on importance scores, yet typically rely on instantaneous gradients that capture only local sensitivity, overlooking non-local, pathwise effects within the same layer, which yields unstable and biased scores. To address this limitation, we introduce IGU-LoRA, an adaptive-rank LoRA that (i) computes within-layer Integrated Gradients (IG) sensitivities and aggregates them into a layer-level score for rank allocation, and (ii) applies an uncertainty-aware scheme using exponential moving averages with deviation tracking to suppress noisy updates and calibrate rank selection. Theoretically, we prove an upper bound on the composite trapezoidal rule approximation error for parameter-space IG under a pathwise Hessian-Lipschitz condition, which informs the quadrature budget. Across diverse tasks and architectures, IGU-LoRA consistently outperforms strong PEFT baselines at matched parameter budgets, improving downstream accuracy and robustness. Ablations confirm the contributions of pathwise within-layer sensitivity estimates and uncertainty-aware selection to effective rank allocation.

Large Language Models (LLMs) currently respond to every prompt. However, they can produce incorrect answers when they lack knowledge or capability -- a problem known as hallucination. We instead propose post-training an LLM to generate content only when confident in its correctness and to otherwise (partially) abstain. Specifically, our method, HALT, produces capability-aligned post-training data that encodes what the model can and cannot reliably generate. We generate this data by splitting responses of the pretrained LLM into factual fragments (atomic statements or reasoning steps), and use ground truth information to identify incorrect fragments. We achieve capability-aligned finetuning responses by either removing incorrect fragments or replacing them with "Unsure from Here" -- according to a tunable threshold that allows practitioners to trade off response completeness and mean correctness of the response's fragments. We finetune four open-source models for biography writing, mathematics, coding, and medicine with HALT for three different trade-off thresholds. HALT effectively trades off response completeness for correctness, increasing the mean correctness of response fragments by 15% on average, while resulting in a 4% improvement in the F1 score (mean of completeness and correctness of the response) compared to the relevant baselines. By tuning HALT for highest correctness, we train a single reliable Llama3-70B model with correctness increased from 51% to 87% across all four domains while maintaining 53% of the response completeness achieved with standard finetuning.

Multimodal large language models (MLLMs) have achieved remarkable progress in vision–language reasoning, yet they remain vulnerable to hallucination, where generated content deviates from the visual evidence. Existing mitigation strategies either demand costly supervision during training or introduce additional latency at inference. Recent vision-enhancement methods attempt to address this by reinforcing visual tokens during decoding, but they typically inject all tokens indiscriminately, leading to interference from background regions and distracting the model from critical cues. To overcome this challenge, we propose an Adaptive vIsual Reinforcement framework for MLLMs, dubbed as AIR. AIR consists of two main components: prototype-based token reduction, which condenses the large pool of visual tokens into a compact subset to suppress redundancy, and OT-guided patch reinforcement, which quantifies the alignment between hidden state and patch embeddings to selectively integrate the most consistent patches into the feed-forward layers. As a result, AIR enhances the model’s reliance on salient visual information and effectively mitigates hallucination. Extensive experiments across representative MLLMs demonstrate that AIR substantially reduces hallucination while preserving general capabilities, establishing it as an effective and independent solution for building reliable MLLMs.

Learning from Label Proportions~(LLP) aims to use bags of instances associated with the proportions of each label within the bag to learn an instance-level classifier. Proportion matching is a widely used strategy that aligns the average model outputs of all instances in a bag with the label proportions in order to induce the classifier. However, simply fitting the label proportions does not encourage discriminative instance-level predictions and may cause over-smoothing problems, resulting in poor classification performance. In this paper, we propose a novel LLP approach that can mitigate the over-smoothing problems with theoretical guarantees. Rather than fitting the label proportions directly, we treat them as targets for an auxiliary proportional value classification task to induce the target classifier. Our approach only requires the incorporation of an aggregation function after the classification layer. We also introduce an efficient computational approach with a divide-and-conquer strategy. Extensive experiments on various benchmark datasets and under different bag-generation strategies demonstrate that our approach achieves superior performance compared with state-of-the-art LLP methods. The code is publicly available at https://github.com/TianhaoMa5/ICLR2026_LLP-PVC.

Vision-Language Models (VLMs) achieve strong results on multimodal tasks such as visual question answering, yet they can still fail even when the correct visual evidence is present. In this work, we systematically investigate whether these failures arise from not perceiving the evidence or from not leveraging it effectively. By examining layer-wise attention dynamics, we find that shallow layers focus primarily on text, while deeper layers sparsely but reliably attend to localized evidence regions. Surprisingly, VLMs often perceive the visual evidence when outputting incorrect answers, a phenomenon we term "seeing but not believing" that widely exists in major VLM families. Building on this, we introduce an inference-time intervention that highlights deep-layer evidence regions through selective attention-based masking. It requires no training and consistently improves accuracy across multiple families, including LLaVA, Qwen, Gemma, and InternVL. These results show that VLMs encode reliable evidence internally but under-utilize it, and that making such signals explicit can bridge the gap between perception and reasoning, advancing the diagnostic understanding and reliability of VLMs.

Recent years have witnessed increasing interests in tackling heteroscedastic noise in bandits and reinforcement learning \citep{zhou2021nearly, zhao2023variance, jia2024does, pacchiano2025second}. In these works, the cumulative variance of the noise $\Lambda = \sum_{t=1}^T \sigma_t^2$, where $\sigma_t^2$ is the variance of the noise at round $t$, has been used to characterize the statistical complexity of the problem, yielding simple regret bounds of order $\tilde{\mathcal O}(d \sqrt{\Lambda / T^2})$ for linear bandits with heteroscedastic noise \citep{zhou2021nearly, zhao2023variance}. However, with a closer look, $\Lambda$ remains the same order even if the noise is close to zero at half of the rounds, which indicates that the $\Lambda$-dependence is not optimal. In this paper, we revisit the linear bandit problem with heteroscedastic noise. We consider the setting where the action set is fixed throughout the learning process. We propose a novel variance-adaptive algorithm VAEE (Variance-Aware Exploration with Elimination) for large action set, which actively explores actions that maximizes the information gain among a candidate set of actions that are not eliminated. With the active-exploration strategy, we show that VAEE achieves a *simple regret* with a nearly *harmonic-mean* dependent rate, i.e. $\tilde{\mathcal O}\Big(d\Big[\sum_{t = 1}^T \frac{1}{\sigma_t^2} - \sum_{i = 1}^{\tilde{O}(d)} \frac{1}{[\sigma^{(i)}]^2} \Big]^{-\frac{1}{2}}\Big)$ where $d$ is the dimension of the feature space and $\sigma^{(i)}$ is the $i$-th smallest variance among $\\{\sigma_t\\}_{t=1}^T$. For finitely many actions, we propose a variance-aware variant of G-optimal design based exploration, which achieves a $\tilde {\mathcal O}$ $\bigg(\sqrt{d \log |\mathcal A| }\Big[ \sum\_{t = 1}\^T \frac{1}{\sigma\_t\^2}- \sum\_{i = 1}^{\tilde{O}(d)} \frac{1}{[\sigma^{(i)}]^2} \Big]^{-\frac{1}{2}}\bigg)$ simple regret bound. We also establish a nearly matching lower bound for the fixed action set setting indicating that \emph{harmonic-mean} dependent rate is unavoidable. To the best of our knowledge, this is the first work that breaks the $\sqrt{\Lambda}$ barrier for linear bandits with heteroscedastic noise.

Linear attention mechanisms have emerged as efficient alternatives to Softmax attention, exhibiting steady improvements in language modeling capabilities driven by increasingly sophisticated designs for decay matrices—though their structural complexity has typically been limited to the Diagonal-Plus-Rank-1 level. To further advance the understanding and capabilities of linear attention via more complex decay structures, this work makes two primary contributions: (1) We propose the HDLA linear attention mechanism, which utilizes efficient matrix decomposition to achieve a Diagonal-Plus-Rank-2 structure, thereby extending the decay matrix to a broader, more expressive, rank-enhanced and structured class. (2) We propose a more general chunk-wise parallel algorithm that accommodates both diagonal-plus-rank-$r_{ab}$ decay structure and key-value outer products of rank $r_{kv}$, thus providing a versatile foundation for future research. Comprehensive experiments demonstrate that, compared to linear attention baselines, HDLA sets new SOTA results on language modeling and retrieval tasks at 2.8B parameter scale, delivers at most 80\% and 58.2\% performance gains over baselines on retrieval-based MQAR and RULER tasks, and achieves an average score improvement of 4.39–7.66 on the synthetic MAD benchmark, respectively. Our proposed HDLA model, as well as the rank-generalized chunk-wise parallel algorithm, together provide a versatile algorithmic foundation and promising research prospects for the design of rank-enhanced, structured linear attention mechanisms.

Collecting web data to train deep models has become increasingly common, raising concerns about unauthorized data usage. To mitigate this issue, unlearnable examples introduce imperceptible perturbations into data, preventing models from learning effectively. However, existing methods typically rely on deep neural networks as surrogate models for perturbation generation, resulting in significant computational costs. In this work, we propose Perturbation-Induced Linearization (PIL), a computationally efficient yet effective method that generates perturbations using only linear surrogate models. PIL achieves comparable or better performance than existing surrogate-based methods while reducing computational time dramatically. We further reveal a key mechanism underlying unlearnable examples: inducing linearization to deep models, which explains why PIL can achieve competitive results in a very short time. Beyond this, we provide an analysis about the property of unlearnable examples under percentage-based partial perturbation. Our work not only provides a practical approach for data protection but also offers insights into what makes unlearnable examples effective.

Generating long-form, multi-speaker conversational audio like podcasts poses significant challenges for traditional Text-to-Speech (TTS) systems, particularly in scalability, speaker consistency, and natural turn-taking. We present VibeVoice , a novel model designed to synthesize expressive, long-form speech with multiple speakers in a zero-shot manner. A core component of our approach is the continuous speech tokenizers operating at an ultra-low frame rate of 7.5. This tokenizer effectively preserves audio fidelity while significantly boosting computational efficiency for processing long sequences. To facilitate training on authentic conversational dynamics, we have developed an annotation pipeline that generates pseudo transcriptions and turn-taking labels for extensive podcast data. Leveraging this data and our efficient tokenizer, VibeVoice employs the next-token diffusion framework. This enables VibeVoice to: (1) synthesize long-form speech (up to 30 minutes) with up to 4 speakers, surpassing the typical 1-2 speaker limits of many prior models; and (2) achieve a high degree of naturalness in turn-taking, pacing, and the rendition of subtle non-lexical cues (such as breaths and lip smacks), which are crucial for listener immersion and capturing the authentic vibe of expressive conversations.

Large Language Model (LLM) agents are increasingly applied to complex, multi-step tasks that require interaction with diverse external tools across various domains. However, current LLM agent tool planning methods typically rely on greedy, reactive tool selection strategies that lack foresight and fail to account for inter-tool dependencies. In this paper, we present ToolTree, a novel Monte-Carlo tree search-inspired planning paradigm for tool planning. ToolTree explores possible tool usage trajectories using a dual-stage LLM evaluation and bidirectional pruning mechanism that enables the agent to make informed, adaptive decisions over extended tool-use sequences while pruning less promising branches before and after the tool execution. Empirical evaluations across both open-set and closed-set tool planning tasks on 4 benchmarks demonstrate that ToolTree consistently improves performance while keeping the highest efficiency, achieving an average gain of around 10\% compared to the state-of-the-art planning paradigm.

The edifice of native Vision-Language Models (VLMs) has emerged as a rising contender to typical modular VLMs, shaped by evolving model architectures and training paradigms. Yet, two lingering clouds cast shadows over its widespread exploration and promotion: (-) What fundamental constraints set native VLMs apart from modular ones, and to what extent can these barriers be overcome? (-) How to make research in native VLMs more accessible and democratized, thereby accelerating progress in the field. In this paper, we clarify these challenges and outline guiding principles for constructing native VLMs. Specifically, one native VLM primitive should: (i) effectively align pixel and word representations within a shared semantic space; (ii) seamlessly integrate the strengths of formerly separate vision and language modules; (iii) inherently embody various cross-modal properties that support unified vision-language encoding, aligning, and reasoning. Hence, we launch NEO, a novel family of native VLMs built from first principles, greatly narrowing the gap with top-tier modular counterparts across diverse real-world scenarios. With 390M image-text examples, NEO efficiently develops visual perception from scratch while mitigating vision-language conflicts inside a dense and monolithic model crafted from our elaborate primitives. We position NEO as a cornerstone for scalable and powerful native VLM development, paired with a rich set of reusable components that foster a cost-effective and extensible ecosystem.

Recent frontier models employ long-chain-of-thought reasoning to explore solution spaces in context and achieve stronger performance. While many works study distillation to build smaller yet capable models, most focus on English and little is known about language-specific reasoning. To bridge this gap, we first introduce **Language-Mixed CoT**, a reasoning schema that switches between English and a target language, using English as an anchor to excel in reasoning while minimizing translation artifacts. As a Korean case study, we curate **Yi-Sang-HQ**: 5.79M native-Korean prompts from web Q&A, exams, STEM, and code; 3.7M long reasoning traces generated from Qwen3-32B; and a targeted 260k high-yield subset. We train nine models (4B–35B) across six families (Qwen2.5, Llama-3.1, Gemma-3, etc). Our best model, **KO-REAson-35B**, achieves state-of-the-art performance, with the highest overall average score ($64.0_{\pm2.5}$), ranking first on 5/9 benchmarks and second on the remainder. Smaller and mid-sized models also benefit substantially, with an average improvement of $+18.6$ points across the evaluated nine benchmarks. Ablations show **Language-Mixed CoT** is more effective than monolingual CoT, indicating that reasoning patterns can be engineered to improve non-English performance. We release our data-curation pipeline, evaluation system, datasets, and models to advance research on language-specific reasoning.

In a multi-follower Bayesian Stackelberg game, a leader plays a mixed strategy over $L$ actions to which $n\ge 1$ followers, each having one of $K$ possible private types, best respond. The leader's optimal strategy depends on the distribution of the followers' private types. We study an online learning version of this problem: a leader interacts for $T$ rounds with $n$ followers with types sampled from an unknown distribution every round. The leader's goal is to minimize regret, defined as the difference between the cumulative utility of the optimal strategy and that of the actually chosen strategies. We design learning algorithms for the leader under different feedback settings. Under type feedback, where the leader observes the followers' types after each round, we design algorithms that achieve $O\big(\sqrt{\min(L\log(nKA T), ~ nK ) \cdot T} \big)$ regret for independent type distributions and $O\big(\sqrt{\min(L\log(nKA T), ~ K^n ) \cdot T} \big)$ regret for general type distributions. Interestingly, those bounds do not grow with $n$ at a polynomial rate. Under action feedback, where the leader only observes the followers' actions, we design algorithms with $O( \min(\sqrt{ n^L K^L A^{2L} L T \log T}, ~ K^n\sqrt{ T } \log T ) )$ regret. We also provide a lower bound of $\Omega(\sqrt{\min(L, ~ nK)T})$, almost matching the type-feedback upper bounds.

Preference optimization is crucial for aligning large language models (LLMs) with human values and intentions. A significant challenge in this process is the distribution mismatch between pre-collected offline preference data and the evolving model policy. Existing methods attempt to reduce this gap using static heuristics or decoupled online sampling strategies, but they often fail to adapt to the model's dynamic learning state. To bridge this gap, we propose Meta-Weighted Adaptive Preference Optimization (MetaAPO), a novel framework that dynamically couples data generation with model training. MetaAPO employs a lightweight meta-learner, as an "alignment gap estimator", to evaluate the potential benefits of on-policy sampling in relation to offline data. This guides targeted online generation and assigns sample-wise meta-weights to the optimization objective, dynamically balancing the quality and distribution of online and offline data. Experiments on AlpacaEval 2, Arena-Hard and MT-Bench demonstrate that MetaAPO consistently outperforms existing preference optimization approaches across various settings, while reducing 42% in online annotation costs.

Fake news undermines societal trust and decision-making across politics, economics, health, and international relations, and in extreme cases threatens human lives and societal safety. Because fake news reflects region-specific political, social, and cultural contexts and is expressed in language, evaluating the risks of large language models (LLMs) requires a multi-lingual and regional perspective. Malicious users can bypass safeguards through jailbreak attacks, inducing LLMs to generate fake news. However, no benchmark currently exists to systematically assess attack resilience across languages and regions. Here, we propose JailNewsBench, the first benchmark for evaluating LLM robustness against jailbreak-induced fake news generation. JailNewsBench spans 34 regions and 22 languages, covering 8 evaluation sub-metrics through LLM-as-a-Judge and 5 jailbreak attacks, with approximately 300k instances. Our evaluation of 9 LLMs reveals that the maximum attack success rate (ASR) reached 86.3% and the maximum harmfulness score was 3.5 out of 5. Notably, for English and U.S.-related topics, the defensive performance of typical multi-lingual LLMs was significantly lower than for other regions, highlighting substantial imbalances in safety across languages and regions. In addition, our analysis shows that coverage of fake news in existing safety datasets is limited and less well defended than major categories such as toxicity and social bias. Our dataset and code are available at https://github.com/kanekomasahiro/jailnewsbench.

Implicit Chain-of-Thought (CoT) methods offer a token-efficient alternative to explicit CoT reasoning in Large Language Models (LLMs), but a persistent performance gap has limited their adoption. We identify a core latent instability issue when scaling the computational budget of implicit CoT: as the number of reasoning tokens increases, training often becomes unstable and collapses. Our analysis shows that this instability arises from latent representations becoming homogeneous and losing semantic diversity, caused by insufficient step-level supervision in current implicit CoT methods. To address this, we propose SIM-CoT, a plug-and-play training module that introduces step-level supervision to stabilize and enrich the latent reasoning space. SIM-CoT employs an auxiliary decoder during training to align each implicit token with its corresponding explicit reasoning step, ensuring latent states capture distinct and meaningful information. The auxiliary decoder is removed at inference, preserving the efficiency of implicit CoT with no added overhead. It also provides interpretability by projecting each latent token onto an explicit reasoning vocabulary, enabling per-step visualization and diagnosis. SIM-CoT significantly improves both in-domain accuracy and out-of-domain stability of implicit CoT methods, boosting Coconut by +8.2\% on GPT-2 and CODI by +3.0\% on LLaMA-3.1 8B. It further surpasses the explicit CoT baseline on GPT-2 by 2.1\% with 2.3$\times$ greater token efficiency, while closing the performance gap on larger models like LLaMA-3.1 8B.

High-quality scientific illustrations are crucial for effectively communicating complex scientific and technical concepts, yet their manual creation remains a well-recognized bottleneck in both academia and industry. We present FigureBench, the first large-scale benchmark for generating scientific illustrations from long-form scientific texts. It contains 3,300 high-quality scientific text–figure pairs, covering diverse text-to-illustration tasks from scientific papers, surveys, blogs, and textbooks. Moreover, we propose AutoFigure, an agentic framework that automatically generates high-quality scientific illustrations based on long-form scientific text. Specifically, before rendering the final result, AutoFigure engages in extensive thinking, recombination, and validation to produce a layout that is both structurally sound and aesthetically refined, outputting a scientific illustration that achieves both structural completeness and aesthetic appeal. Leveraging the high-quality data from FigureBench, we conduct extensive experiments to test the performance of AutoFigure against various baseline methods. The results demonstrate that Autofigure consistently surpasses all baseline methods, producing publication-ready scientific illustrations.

We make two contributions to the problem of estimating the $L_1$ calibration error of a binary classifier from a finite dataset. First, we provide an upper bound for any classifier where the calibration function has bounded variation. Second, we provide a method of modifying any classifier so that its calibration error can be upper bounded efficiently without significantly impacting classifier performance and without any restrictive assumptions. All our results are non-asymptotic and distribution-free. We conclude by providing advice on how to measure calibration error in practice. Our methods yield practical procedures that can be run on real-world datasets with modest overhead.

Visual Prompt Tuning (VPT) enables effective adaptation of a frozen Vision Transformer (ViT) to downstream tasks by inserting a small number of learnable prompt tokens into the token sequence at each layer. However, we observe that existing VPT variants often suffer from unstable training dynamics, characterized by gradient oscillations. A closer layer-wise analysis reveals that shallow-layer prompts tend to stagnate early, while deeper-layer prompts exhibit high-variance oscillations, leading to a cross-layer mismatch. These issues contribute to slower convergence and degraded final performance. To address these challenges, we propose the Prompt-Agnostic Evolution ($\mathtt{PAE}$) method, which can strengthen vision prompt tuning by explicitly modeling the dynamics of learnable prompts. From a frequency-domain perspective, we initialize prompts in a task-aware direction by uncovering and propagating frequency shortcut patterns that the backbone inherently exploits for recognition. To ensure coherent evolution across layers, we further employ a shared Koopman operator, which imposes a global linear transformation rather than uncoordinated, layer-specific updates. Finally, inspired by Lyapunov stability theory, we introduce a regularizer that constrains error amplification during evolution. Extensive experiments demonstrate that using $\mathtt{PAE}$ with VPT variants not only accelerates convergence with an average 1.41$\times$ speedup but also yields 1–3% gains on 25 datasets with multi downstream tasks. Beyond performance, $\mathtt{PAE}$ remains prompt-agnostic and lightweight, and it integrates seamlessly with diverse VPT variants without backbone modification or inference-time changes, providing a practical and scalable solution for advancing prompt tuning.

While massively both scaling data and models have become central in NLP and 2D vision, their benefits for 3D point cloud understanding remain limited. We study the initial step of 3D point cloud scaling under a realistic regime: large-scale multi-dataset joint training for 3D semantic segmentation, with no dataset labels available at inference time. Point clouds arise from a wide range of sensors (e.g., depth cameras, LiDAR) and scenes (e.g., indoor, outdoor), yielding heterogeneous scanning patterns, sampling densities, and semantic biases; naively mixing such datasets degrades standard models. We introduce **Point-MoE**, a Mixture-of-Experts design that expands capacity through sparsely activated expert MLPs and a lightweight top-$k$ router, allowing tokens to select specialized experts without requiring dataset supervision. Trained jointly on a diverse mix of indoor and outdoor datasets and evaluated on seen datasets and in zero-shot settings, Point-MoE outperforms prior methods without using dataset labels for either training or inference. This outlines a scalable path for 3D perception: letting the model discover structure in heterogeneous 3D data rather than imposing it via manual curation or dataset-specific heuristics.

AdamW has long been the dominant optimizer in language model pretraining, despite numerous claims that alternative optimizers offer 1.4 to 2$\times$ speedup. We posit that two methodological shortcomings have obscured fair comparisons and hindered practical adoption: (i) unequal hyperparameter tuning and (ii) limited or misleading evaluation setups. To address these two issues, we conduct a systematic study of ten deep learning optimizers across four model scales (0.1B–1.2B parameters) and data-to-model ratios (1--8$\times$ the Chinchilla optimum). We find that fair and informative comparisons require rigorous hyperparameter tuning and evaluations across a range of model scales and data-to-model ratios, performed at the end of training. First, optimal hyperparameters for one optimizer may be suboptimal for another, making blind hyperparameter transfer unfair. Second, the actual speedup of many proposed optimizers over well-tuned baselines is lower than claimed and decreases with model size to only 1.1$\times$ for 1.2B parameter models. Thirdly, comparing intermediate checkpoints before reaching the target training budgets can be misleading, as rankings between two optimizers can flip during training due to learning rate decay. Through our thorough investigation, we find that all the fastest optimizers such as Muon and Soap, use matrices as preconditioners --- multiplying gradients with matrices rather than entry-wise scalars. However, the speedup of matrix-based optimizers is inversely proportional to model scale, decreasing from 1.4$\times$ over AdamW for 0.1B parameter models to merely 1.1$\times$ for 1.2B parameter models.

Large Language Models (LLMs) have achieved remarkable success in software engineering tasks when trained with executable runtime environments, particularly in resolving GitHub issues. However, such runtime environments are often unavailable in other domains, especially cybersecurity, where challenge configurations and execution contexts are ephemeral or restricted. We present Cyber-Zero, the first runtime-free framework for synthesizing high-quality agent trajectories to train cybersecurity LLMs. Cyber-Zero leverages publicly available CTF writeups and employs persona-driven LLM simulation to reverse-engineer runtime behaviors and generate realistic, long-horizon interaction sequences without actual environments. Using trajectories synthesized by Cyber-Zero, we train LLM-based agents that achieve up to 13.1% absolute performance gains over baseline models on three prominent CTF benchmarks: InterCode-CTF, NYU CTF Bench, and Cybench. Our best model, Cyber-Zero-32B, establishes new state-of-the-art performance among open-weight models, matching the capabilities of proprietary systems like DeepSeek-V3-0324 and Claude-3.5-Sonnet while offering superior cost-effectiveness, and demonstrating that runtime-free trajectory synthesis can effectively democratize the development of state-of-the-art cybersecurity agents.

We present an algorithm for the empirical group distributionally robust (GDR) least squares problem. Given $m$ groups, a parameter vector in $\mathbb{R}^d$, and stacked design matrices and responses $\mathbf{A}$ and $\bm{b}$, our algorithm obtains a $(1+\varepsilon)$-multiplicative optimal solution using $\widetilde{O}(\min\{\mathsf{rank}(\mathbf{A}),m\}^{1/3}\varepsilon^{-2/3})$ linear-system-solves of matrices of the form $\mathbf{A}^{\top}\mathbf{B}\mathbf{A}$ for block-diagonal $\mathbf{B}$. Our technical methods follow from a recent geometric construction, block Lewis weights, that relates the empirical GDR problem to a carefully chosen least squares problem and an application of accelerated proximal methods. Our algorithm improves over known interior point methods for moderate accuracy regimes and matches the state-of-the-art guarantees for the special case of $\ell_{\infty}$ regression. We also give algorithms that smoothly interpolate between minimizing the average least squares loss and the distributionally robust loss.

Large Language Models (LLMs) have brought about a significant change in the field of artificial intelligence, where they have transitioned in scope from being specialized research tools to common resources that drive the next generation of software. With increasing model parameters and training data, LLMs demonstrate new abilities in reasoning, code generation, and solving complex problems that were once considered unattainable. However, scaling these models effectively for long-context applications uniquely poses a challenge. This is primarily due to the inherent limitations of the self-attention mechanism, which has quadratic time complexity. This quadratic bottleneck hinders applications for long documents, high-resolution images, and large codebases, among others. However, what is interesting to observe is that effectively only a few parameters are used in token computation, and most calculations are sparse. Hence, sparsity emerges as an effective solution to this problem. Rather than relying on the entire attention matrix, one can utilize an approximate or sparse version of attention to achieve almost the same results much faster. The backbone of this approach is the idea that tokens do not require the entire context; they only need local context, and thus, most of the computation carried out is wasteful. In this blog, we analyze the types of attention patterns that emerge and how to use them to our advantage for faster and efficient LLMs.

We consider the probabilistic simplex-constrained sparse recovery problem. The commonly used Lasso-type penalty for promoting sparsity is ineffective in this context since it is a constant within the simplex. Despite this challenge, fortunately, simplex constraint itself brings a self-regularization property, i.e., the empirical risk minimizer without any sparsity-promoting procedure obtains the usual Lasso-type estimation error. Moreover, we analyze the iterates of a projected gradient descent method and show its convergence to the ground truth sparse solution in the geometric rate until a satisfied statistical precision is attained. Although the estimation error is statistically optimal, the resulting solution is usually more dense than the sparse ground truth. To further sparsify the iterates, we propose a method called PERMITS via embedding a tail screening procedure, i.e., identifying negligible components and discarding them during iterations, into the projected gradient descent method. Furthermore, we combine tail screening and the special information criterion to balance the trade-off between fitness and complexity. Theoretically, the proposed PERMITS method can exactly recover the ground truth support set under mild conditions and thus obtain the oracle property. We demonstrate the statistical and computational efficiency of PERMITS with both synthetic and real data. The implementation of the proposed method can be found in https://github.com/abess-team/PERMITS.

Online alignment (e.g., GRPO) is generally more performant than offline alignment (e.g., DPO)---but why? Drawing on prospect theory from behavioral economics, we propose a human-centric explanation. We prove that online on-policy sampling better approximates the human-perceived distribution of what the model can produce, and PPO/GRPO-style clipping---originally introduced to just stabilize training---recovers a perceptual bias in how humans perceive probability. In this sense, PPO/GRPO act as perceptual losses already. Our theory further suggests that the online/offline dichotomy is itself incidental to maximizing human utility, since we can achieve the same effect by selectively training on any data in a manner that mimics human perception, rather than restricting ourselves to online on-policy data. Doing so would allow us to post-train more quickly, cheaply, and flexibly without sacrificing performance. To this end, we propose a design pattern that explicitly incorporates perceptual distortions of probability into objectives like DPO/KTO/GRPO, creating $\textit{humanline variants}$ of them. Surprisingly, we find that these humanline variants, even when trained with offline off-policy data, can match the performance of their online counterparts on both verifiable and unverifiable tasks.

Computer-use agents (CUAs) hold the promise of performing a wide variety of general tasks, but current evaluations have primarily focused on simple scenarios. It therefore remains unclear whether such generalist agents can automate more sophisticated and specialized work such as software engineering (SWE). To investigate this, we introduce Programming with Pixels (PwP), the first comprehensive computer-use environment for software engineering, where agents visually control an IDE to perform diverse software engineering tasks. To enable holistic evaluation, we also introduce PwP-Bench, a benchmark of 15 existing and new software-engineering tasks spanning multiple modalities, programming languages, and skillsets. We perform an extensive evaluation of state-of-the-art open-weight and closed-weight CUAs and find that when interacting purely visually, they perform significantly worse than specialized coding agents. However, when the same CUAs are given direct access to just two APIs—file editing and bash operations—performance jumps, often reaching the levels of specialized agents despite having a task-agnostic design. Furthermore, when given access to additional IDE tools via text APIs, all models show further gains. Our analysis shows that current CUAs fall short mainly due to limited visual grounding and the inability to take full advantage of the rich environment, leaving clear room for future improvements. PwP establishes software engineering as a natural domain for benchmarking whether generalist computer-use agents can reach specialist-level performance on sophisticated tasks.