ICLR 2017

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

iclr2017:conference_posters [2017/03/27 17:31]
hugo created
iclr2017:conference_posters [2017/04/23 09:26] (current)
hugo
Line 1: Line 1:
-====Conference Poster Sessions====+======Conference Poster Sessions======
  
-Below are the Conference Track papers presented at each of the poster sessions (on Monday, Tuesday or Wednesday, in the morning or evening). To find a paper, look for the poster with the corresponding number.+Below are the Conference Track papers presented at each of the poster sessions (on Monday, Tuesday or Wednesday, in the morning or evening). To find a paper, look for the poster with the corresponding number ​in the area dedicated to the Conference Track.
  
-===Monday Morning=== +======Note to the Presenters======= 
-  - Making Neural Programming Architectures Generalize via Recursion +Each poster panel is 2 meters large and 1 meter tall.\\ 
-  - Learning Graphical State Transitions +If neededtape will be provided ​to fix your poster.
-  - Distributed Second-Order Optimization using Kronecker-Factored Approximations +
-  - Normalizing the Normalizers:​ Comparing and Extending Network Normalization Schemes +
-  - Neural Program Lattices +
-  - Diet Networks: Thin Parameters for Fat Genomics +
-  - Unsupervised Cross-Domain Image Generation +
-  - Towards Principled Methods for Training Generative Adversarial Networks +
-  - Recurrent Mixture Density Network for Spatiotemporal Visual Attention +
-  - Paying More Attention ​to Attention: Improving ​the Performance of Convolutional Neural Networks via Attention Transfer +
-  - Pruning Filters for Efficient ConvNets +
-  - Optimization as a Model for Few-Shot Learning +
-  - Understanding deep learning requires rethinking generalization +
-  - On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima +
-  - Recurrent Hidden Semi-Markov Model +
-  - Nonparametric Neural Networks +
-  - Learning to Generate Samples from Noise through Infusion Training +
-  - An Information-Theoretic Framework for Fast and Robust Unsupervised Learning via Neural Population Infomax +
-  - Highway and Residual Networks learn Unrolled Iterative Estimation +
-  - Soft Weight-Sharing for Neural Network Compression +
-  - Snapshot Ensembles: Train 1Get M for Free +
-  - Towards a Neural Statistician +
-  - Learning Curve Prediction with Bayesian Neural Networks +
-  - Learning End-to-End Goal-Oriented Dialog +
-  - Multi-Agent Cooperation and the Emergence of (Natural) Language +
-  - Efficient Vector Representation for Documents through Corruption +
-  - Improving Neural Language Models with a Continuous Cache +
-  - Program Synthesis for Character Level Language Modeling +
-  - Tracking the World State with Recurrent Entity Networks +
-  - Reinforcement Learning with Unsupervised Auxiliary Tasks +
-  - Neural Architecture Search with Reinforcement Learning +
-  - Sample Efficient Actor-Critic with Experience Replay +
-  - Learning to Act by Predicting the Future+
  
-===Monday Afternoon=== 
-  - Neuro-Symbolic Program Synthesis 
-  - Generative Models and Model Criticism via Optimized Maximum Mean Discrepancy 
-  - Trained Ternary Quantization 
-  - DSD: Dense-Sparse-Dense Training for Deep Neural Networks 
-  - A Compositional Object-Based Approach to Learning Physical Dynamics 
-  - Multilayer Recurrent Network Models of Primate Retinal Ganglion Cells 
-  - Improving Generative Adversarial Networks with Denoising Feature Matching 
-  - Transfer of View-manifold Learning to Similarity Perception of Novel Objects 
-  - What does it take to generate natural textures? 
-  - Emergence of foveal image sampling from learning to attend in visual scenes 
-  - PixelCNN++: A PixelCNN Implementation with Discretized Logistic Mixture Likelihood and Other Modifications 
-  - Learning to Optimize 
-  - Training Compressed Fully-Connected Networks with a Density-Diversity Penalty 
-  - Optimal Binary Autoencoding with Pairwise Correlations 
-  - On the Quantitative Analysis of Decoder-Based Generative Models 
-  - Learning to Remember Rare Events 
-  - Transfer Learning for Sequence Tagging with Hierarchical Recurrent Networks 
-  - Capacity and Learnability in Recurrent Neural Networks 
-  - Deep Learning with Dynamic Computation Graphs 
-  - Exploring Sparsity in Recurrent Neural Networks 
-  - Structured Attention Networks 
-  - Zoneout: Regularizing RNNs by Randomly Preserving Hidden Activations 
-  - Variational Lossy Autoencoder 
-  - Learning to Query, Reason, and Answer Questions On Ambiguous Texts 
-  - Deep Biaffine Attention for Neural Dependency Parsing 
-  - A Compare-Aggregate Model for Matching Text Sequences 
-  - Data Noising as Smoothing in Neural Network Language Models 
-  - Neural Variational Inference For Topic Models 
-  - Words or Characters? Fine-grained Gating for Reading Comprehension 
-  - Q-Prop: Sample-Efficient Policy Gradient with An Off-Policy Critic 
-  - Stochastic Neural Networks for Hierarchical Reinforcement Learning 
-  - Learning Invariant Feature Spaces to Transfer Skills with Reinforcement Learning 
-  - Third Person Imitation Learning 
  
-===Tuesday ​Morning=== +<​html><​div id='​monday_morning'></​div></​html>​ 
-  - DeepDSLA Compilation-based Domain-Specific Language for Deep Learning +====Monday ​Morning ​(April 24th, 10:30am to 12:30pm)==== 
-  - SampleRNNAn Unconditional End-to-End Neural Audio Generation Model +C1: Making Neural Programming Architectures Generalize via Recursion\\ 
-  - Deep Probabilistic Programming +C2: Learning ​Graphical State Transitions\\ 
-  - Lie-Access ​Neural ​Turing Machines +C3Distributed Second-Order Optimization using Kronecker-Factored Approximations\\ 
-  - Learning Features of Music From Scratch +C4: Normalizing the Normalizers:​ Comparing and Extending Network Normalization Schemes\\ 
-  - Mode Regularized Generative Adversarial ​Networks +C5: Neural ​Program Lattices\\ 
-  End-to-end Optimized ​Image Compression +C6: Diet Networks: Thin Parameters for Fat Genomics\\ 
-  - Variational Recurrent ​Adversarial ​Deep Domain Adaptation +C7: Unsupervised Cross-Domain ​Image Generation\\ 
-  - Steerable CNNs +C8: Towards Principled Methods for Training Generative ​Adversarial Networks\\ 
-  - Deep Predictive Coding ​Networks ​for Video Prediction and Unsupervised Learning +C9Recurrent Mixture Density Network ​for Spatiotemporal Visual Attention\\ 
-  - PixelVAEA Latent Variable Model for Natural Images +C10: Paying More Attention to Attention: Improving the Performance of Convolutional ​Neural Networks ​via Attention Transfer\\ 
-  - A recurrent neural network without chaos +C11Pruning Filters for Efficient ConvNets\\ 
-  - Outrageously Large Neural Networks: ​The Sparsely-Gated Mixture-of-Experts Layer +C12: Stick-Breaking Variational Autoencoders\\ 
-  - Tree-structured decoding with doubly-recurrent neural networks +C13Identity Matters in Deep Learning\\ 
-  - Introspection:Accelerating Neural Network Training By Learning ​Weight Evolution +C14On Large-Batch Training ​for Deep Learning: Generalization Gap and Sharp Minima\\ 
-  - HyperbandBandit-Based Configuration Evaluation ​for Hyperparameter Optimization +C15: Recurrent ​Hidden Semi-Markov Model\\ 
-  - Quasi-Recurrent Neural Networks +C16: Nonparametric ​Neural Networks\\ 
-  - Attend, Adapt and TransferAttentive Deep Architecture for Adaptive Transfer ​from multiple sources in the same domain +C17Learning to Generate Samples ​from Noise through Infusion Training\\ 
-  A Baseline ​for Detecting Misclassified ​and Out-of-Distribution Examples in Neural Networks +C18: An Information-Theoretic Framework ​for Fast and Robust Unsupervised Learning via Neural ​Population Infomax\\ 
-  Trusting SVM for Piecewise Linear CNNs +C19: Highway and Residual ​Networks ​learn Unrolled Iterative Estimation\\ 
-  - Maximum Entropy Flow Networks +C20: Soft Weight-Sharing ​for Neural Network Compression\\ 
-  - The Concrete DistributionA Continuous Relaxation of Discrete Random Variables +C21: Snapshot Ensembles: Train 1, Get M for Free\\ 
-  - Unrolled Generative Adversarial ​Networks +C22Towards a Neural Statistician\\ 
-  - A Simple but Tough-to-Beat Baseline for Sentence Embeddings +C23: Learning Curve Prediction with Bayesian Neural ​Networks\\ 
-  ​Query-Reduction Networks for Question Answering +C24: Learning End-to-End Goal-Oriented Dialog\\ 
-  Machine Comprehension Using Match-LSTM ​and Answer Pointer +C25: Multi-Agent Cooperation ​and the Emergence of (Natural) Language\\ 
-  - Bidirectional Attention Flow for Machine Comprehension +C26: Efficient Vector Representation ​for Documents through Corruption\\ 
-  - Dynamic Coattention Networks For Question Answering +C27: Improving ​Neural ​Language Models with a Continuous Cache\\ 
-  - Multi-view Recurrent ​Neural ​Acoustic Word Embeddings +C28: Program Synthesis ​for Character Level Language Modeling\\ 
-  - Episodic Exploration ​for Deep Deterministic Policies for StarCraft Micromanagement +C29: Tracking the World State with Recurrent Entity Networks\\ 
-  - Training Agent for First-Person Shooter Game with Actor-Critic Curriculum ​Learning +C30: Reinforcement ​Learning ​with Unsupervised Auxiliary Tasks\\ 
-  - Generalizing Skills ​with Semi-Supervised ​Reinforcement Learning +C31: Neural Architecture Search ​with Reinforcement Learning\\ 
-  Improving Policy Gradient ​by Exploring Under-appreciated Rewards+C32: Sample Efficient Actor-Critic with Experience Replay\\ 
 +C33: Learning to Act by Predicting the Future\\
  
-===Tuesday ​Afternoon=== +<​html><​div id='​monday_afternoon'></​div></​html>​ 
-  Sigma Delta Quantized Networks +====Monday ​Afternoon ​(April 24th, 4:30pm to 6:30pm)==== 
-  - PaleoA Performance ​Model for Deep Neural Networks +C1: Neuro-Symbolic Program Synthesis\\ 
-  - DeepCoderLearning to Write Programs +C2Generative Models and Model Criticism via Optimized Maximum Mean Discrepancy\\ 
-  - Topology and Geometry of Deep Rectified Network Optimization Landscapes +C3Trained Ternary Quantization\\ 
-  - Incremental Network QuantizationTowards Lossless CNNs with Low-precision Weights +C4DSD: Dense-Sparse-Dense Training ​for Deep Neural Networks\\ 
-  ​Learning to Perform Physics Experiments via Deep Reinforcement Learning +C5: A Compositional Object-Based Approach to Learning Physical Dynamics\\ 
-  - Decomposing Motion and Content ​for Natural Video Sequence Prediction +C6: Multilayer Recurrent Network Models of Primate Retinal Ganglion Cells\\ 
-  - Calibrating Energy-based Generative Adversarial Networks +C7Improving ​Generative Adversarial Networks ​with Denoising Feature Matching\\ 
-  - Pruning Convolutional ​Neural Networks ​for Resource Efficient Inference +C8: Transfer of View-manifold ​Learning ​to Similarity Perception of Novel Objects\\ 
-  Incorporating long-range consistency in CNN-based texture generation +C9What does it take to generate natural textures?​\\ 
-  - Lossy Image Compression with Compressive Autoencoders +C10: Emergence ​of foveal image sampling ​from learning to attend in visual scenes\\ 
-  - LR-GANLayered Recursive ​Generative Adversarial Networks ​for Image Generation +C11: PixelCNN++: A PixelCNN Implementation with Discretized Logistic Mixture Likelihood and Other Modifications\\ 
-  Semi-supervised Knowledge Transfer for Deep Learning ​from Private Training Data +C12: Learning ​to Optimize\\ 
-  - Deep Variational Bayes FiltersUnsupervised Learning ​of State Space Models ​from Raw Data +C13: Do Deep Convolutional Nets Really Need to be Deep and Convolutional?​\\ 
-  - Mollifying Networks +C14: Optimal Binary Autoencoding ​with Pairwise Correlations\\ 
-  - beta-VAE: Learning ​Basic Visual Concepts with a Constrained Variational Framework +C15: On the Quantitative Analysis of Decoder-Based Generative Models\\ 
-  - Categorical Reparameterization ​with Gumbel-Softmax +C16: Adversarial machine learning at scale\\ 
-  Online Bayesian ​Transfer Learning for Sequential Data Modeling +C17: Transfer Learning for Sequence ​Tagging with Hierarchical Recurrent Networks\\ 
-  - Latent ​Sequence ​Decompositions +C18: Capacity and Learnability in Recurrent ​Neural Networks\\ 
-  - Density estimation using Real NVP +C19Deep Learning ​with Dynamic ​Computation ​Graphs\\ 
-  ​- ​Recurrent ​Batch Normalization +C20: Exploring Sparsity ​in Recurrent Neural Networks\\ 
-  - SGDRStochastic Gradient Descent ​with Restarts +C21: Structured Attention Networks\\ 
-  - Variable ​Computation in Recurrent Neural Networks +C22: Learning to Repeat: Fine Grained Action Repetition for Deep Reinforcement Learning\\ 
-  ​- ​Deep Variational ​Information Bottleneck +C23: Variational ​Lossy Autoencoder\\ 
-  - A SELF-ATTENTIVE SENTENCE EMBEDDING +C24: Learning to Query, Reason, and Answer Questions On Ambiguous Texts\\ 
-  - TopicRNNA Recurrent ​Neural ​Network with Long-Range Semantic ​Dependency +C25Deep Biaffine Attention for Neural Dependency ​Parsing\\ 
-  Frustratingly Short Attention Spans in Neural Language ​Modeling +C26: A Compare-Aggregate Model for Matching Text Sequences\\ 
-  - Offline Bilingual Word Vectors Without a Dictionary +C27: Data Noising as Smoothing ​in Neural ​Network ​Language ​Models\\ 
-  - LEARNING A NATURAL LANGUAGE INTERFACE WITH NEURAL PROGRAMMER +C28: Neural Variational Inference For Topic Models\\ 
-  Designing ​Neural ​Network Architectures using Reinforcement Learning +C29: Bidirectional Attention Flow for Machine Comprehension\\ 
-  - Metacontrol for Adaptive Imagination-Based Optimization +C30: Q-Prop: Sample-Efficient Policy Gradient with An Off-Policy Critic\\ 
-  - Recurrent Environment Simulators +C31: Stochastic ​Neural ​Networks for Hierarchical ​Reinforcement Learning\\ 
-  - EPOpt: Learning ​Robust Neural Network Policies Using Model Ensembles+C32: Learning Invariant Feature Spaces to Transfer Skills with Reinforcement Learning\\ 
 +C33Third Person Imitation ​Learning\\
  
-===Wednesday ​Morning=== +<​html><​div id='​tuesday_morning'></​div></​html>​ 
-  Deep Multi-task Representation ​Learning: A Tensor Factorisation Approach +====Tuesday ​Morning ​(April 25th, 10:30am to 12:30pm)==== 
-  ​Training deep neural-networks using a noise adaptation layer +C1: DeepDSL: A Compilation-based Domain-Specific Language for Deep Learning\\ 
-  - Delving into Transferable Adversarial Examples and Black-box Attacks +C2: A SELF-ATTENTIVE SENTENCE EMBEDDING\\ 
-  Towards the Limit of Network Quantization +C3: Deep Probabilistic Programming\\ 
-  - Towards Deep Interpretability (MUS-ROVER II): Learning ​Hierarchical Representations ​of Tonal Music +C4: Lie-Access Neural Turing Machines\\ 
-  - Learning to superoptimize programs +C5: Learning ​Features ​of Music From Scratch\\ 
-  - Regularizing CNNs with Locally Constrained Decorrelations +C6: Mode Regularized ​Generative Adversarial Networks\\ 
-  - Generative ​Multi-Adversarial Networks +C7: End-to-end Optimized Image Compression\\ 
-  Visualizing ​Deep Neural Network DecisionsPrediction Difference Analysis +C8: Variational Recurrent Adversarial ​Deep Domain Adaptation\\ 
-  - FractalNetUltra-Deep Neural ​Networks without ​Residuals +C9Steerable CNNs\\ 
-  Faster CNNs with Direct Sparse Convolutions and Guided Pruning +C10: Deep Predictive Coding ​Networks ​for Video Prediction and Unsupervised Learning\\ 
-  ​FILTER SHAPING FOR CONVOLUTIONAL NEURAL NETWORKS +C11: PixelVAE: A Latent Variable Model for Natural Images\\ 
-  The Neural ​Noisy Channel +C12: A recurrent neural network ​without ​chaos\\ 
-  Automatic Rule Extraction from Long Short Term Memory Networks +C13: Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer\\ 
-  Adversarially Learned Inference +C14: Tree-structured decoding with doubly-recurrent neural networks\\ 
-  ​- ​Deep Information Propagation +C15: Introspection:​Accelerating ​Neural ​Network Training By Learning Weight Evolution\\ 
-  Revisiting Classifier Two-Sample Tests +C16: Hyperband: Bandit-Based Configuration Evaluation for Hyperparameter Optimization\\ 
-  - Loss-aware Binarization of Deep Networks +C17: Quasi-Recurrent Neural Networks\\ 
-  - Energy-based ​Generative Adversarial Networks +C18: Attend, Adapt and Transfer: Attentive ​Deep Architecture for Adaptive Transfer from multiple sources in the same domain\\ 
-  Central Moment Discrepancy (CMD) for Domain-Invariant Representation Learning +C19: A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks\\ 
-  Temporal Ensembling ​for Semi-Supervised Learning +C20: Trusting SVM for Piecewise Linear CNNs\\ 
-  On Detecting Adversarial Perturbations +C21: Maximum Entropy Flow Networks\\ 
-  - Identity Matters in Deep Learning +C22: The Concrete Distribution:​ A Continuous Relaxation of Discrete Random Variables\\ 
-  - Adversarial Feature Learning +C23: Unrolled ​Generative Adversarial Networks\\ 
-  - Learning through Dialogue Interactions +C24: A Simple but Tough-to-Beat Baseline for Sentence Embeddings\\ 
-  - Learning to Compose ​Words into Sentences with Reinforcement Learning +C25: Query-Reduction Networks ​for Question Answering\\ 
-  ​Batch Policy Gradient Methods ​for Improving Neural Conversation Models +C26: Machine Comprehension Using Match-LSTM and Answer Pointer\\ 
-  Tying Word Vectors and Word ClassifiersA Loss Framework ​for Language Modeling +C27: Words or Characters? Fine-grained Gating ​for Reading Comprehension\\ 
-  - Geometry of Polysemy +C28: Dynamic Coattention Networks For Question Answering\\ 
-  - PGQCombining policy gradient and Q-learning +C29: Multi-view Recurrent Neural Acoustic ​Word Embeddings\\ 
-  - Reinforcement Learning through Asynchronous Advantage ​Actor-Critic ​on a GPU +C30Episodic Exploration ​for Deep Deterministic Policies for StarCraft Micromanagement\\ 
-  - Learning ​to Navigate in Complex Environments +C31Training Agent for First-Person Shooter Game with Actor-Critic ​Curriculum Learning\\ 
-  Learning and Policy Search in Stochastic Dynamical Systems with Bayesian Neural Networks+C32: Generalizing Skills with Semi-Supervised Reinforcement ​Learning\\ 
 +C33: Improving Policy Gradient by Exploring Under-appreciated Rewards\\
  
-===Wednesday ​Afternoon=== +<​html><​div id='​tuesday_afternoon'></​div></​html>​ 
-  - Learning recurrent representations for hierarchical behavior modeling +====Tuesday ​Afternoon ​(April 25th, 2:00pm to 4:00pm)==== 
-  ​- ​Predicting Medications from Diagnostic Codes with Recurrent Neural Networks +C1: Sigma Delta Quantized Networks\\ 
-  ​- ​Sparsely-Connected Neural Networks: Towards Efficient VLSI Implementation of Deep Neural Networks +C2: Paleo: A Performance Model for Deep Neural Networks\\ 
-  ​- ​HolStep: A Machine Learning Dataset for Higher-order Logic Theorem Proving +C3: DeepCoder: Learning to Write Programs\\ 
-  ​- ​Learning Invariant Representations Of Planar Curves +C4: Topology and Geometry of Deep Rectified Network Optimization Landscapes\\ 
-  ​- ​Entropy-SGD:​ Biasing Gradient Descent Into Wide Valleys +C5: Incremental Network Quantization:​ Towards Lossless CNNs with Low-precision Weights\\ 
-  ​- ​Amortised MAP Inference for Image Super-resolution +C6: Learning to Perform Physics Experiments via Deep Reinforcement Learning\\ 
-  ​- ​Inductive Bias of Deep Convolutional Networks through Pooling Geometry +C7: Decomposing Motion and Content for Natural Video Sequence Prediction\\ 
-  ​- ​Neural Photo Editing with Introspective Adversarial Networks +C8: Calibrating Energy-based Generative Adversarial Networks\\ 
-  ​- ​A Learned Representation For Artistic Style +C9: Pruning Convolutional Neural Networks for Resource Efficient Inference\\ 
-  - Adversarial Machine ​Learning ​at Scale +C10: Incorporating long-range consistency in CNN-based texture generation\\ 
-  Stick-Breaking Variational Autoencoders +C11: Lossy Image Compression with Compressive Autoencoders\\ 
-  ​- ​Support Regularized Sparse Coding and Its Fast Encoder +C12: LR-GAN: Layered Recursive Generative Adversarial Networks for Image Generation\\ 
-  ​- ​Discrete Variational Autoencoders +C13: Semi-supervised Knowledge Transfer for Deep Learning from Private Training Data\\ 
-  Do Deep Convolutional Nets Really Need to be Deep and Convolutional?​ +C14: Deep Variational Bayes Filters: Unsupervised Learning of State Space Models from Raw Data\\ 
-  ​- Efficient Representation of Low-Dimensional Manifolds using Deep Networks +C15: Mollifying Networks\\ 
-  ​- ​Semi-Supervised Classification with Graph Convolutional Networks +C16: beta-VAE: Learning Basic Visual Concepts with a Constrained Variational Framework\\ 
-  ​- ​Understanding Neural Sparse Coding with Matrix Factorization +C17: Categorical Reparameterization with Gumbel-Softmax\\ 
-  ​- ​Tighter bounds lead to improved classifiers +C18: Online Bayesian Transfer Learning for Sequential Data Modeling\\ 
-  ​- ​Why Deep Neural Networks for Function Approximation?​ +C19: Latent Sequence Decompositions\\ 
-  ​- ​Hierarchical Multiscale Recurrent Neural Networks +C20: Density estimation using Real NVP\\ 
-  ​- ​Dropout with Expectation-linear Regularization +C21: Recurrent Batch Normalization\\ 
-  ​- ​HyperNetworks +C22: SGDR: Stochastic Gradient Descent with Restarts\\ 
-  ​- ​Hadamard Product for Low-rank Bilinear Pooling +C23: Variable Computation in Recurrent Neural Networks\\ 
-  ​- ​Adversarial Training Methods for Semi-Supervised Text Classification +C24: Deep Variational Information Bottleneck\\ 
-  ​- ​Fine-grained Analysis of Sentence Embeddings Using Auxiliary Prediction Tasks +C25: SampleRNN: An Unconditional End-to-End Neural Audio Generation Model\\ 
-  ​- ​Pointer Sentinel Mixture Models +C26: TopicRNN: A Recurrent Neural Network with Long-Range Semantic Dependency\\ 
-  ​- ​Reasoning with Memory Augmented Neural Networks for Language Comprehension +C27: Frustratingly Short Attention Spans in Neural Language Modeling\\ 
-  ​- ​Dialogue Learning With Human-in-the-Loop +C28: Offline Bilingual Word Vectors, Orthogonal Transformations and the Inverted Softmax\\ 
-  - Learning to RepeatFine Grained Action Repetition for Deep Reinforcement Learning +C29: LEARNING A NATURAL LANGUAGE INTERFACE WITH NEURAL PROGRAMMER\\ 
-  ​- ​Learning to Play in a Day: Faster Deep Reinforcement Learning by Optimality Tightening +C30: Designing Neural Network Architectures using Reinforcement Learning\\ 
-  ​- ​Learning Visual Servoing with Deep Features and Trust Region Fitted Q-Iteration +C31: Metacontrol for Adaptive Imagination-Based Optimization\\ 
-  ​- ​An Actor-Critic Algorithm for Sequence Prediction+C32: Recurrent Environment Simulators\\ 
 +C33: EPOpt: Learning Robust Neural Network Policies Using Model Ensembles\\ 
 + 
 +<​html><​div id='​wednesday_morning'></​div></​html>​ 
 +====Wednesday Morning (April 26th, 10:30am to 12:​30pm)==== 
 +C1: Deep Multi-task Representation Learning: A Tensor Factorisation Approach\\ 
 +C2: Training deep neural-networks using a noise adaptation layer\\ 
 +C3: Delving into Transferable Adversarial Examples and Black-box Attacks\\ 
 +C4: Towards the Limit of Network Quantization\\ 
 +C5: Towards Deep Interpretability (MUS-ROVER II): Learning Hierarchical Representations of Tonal Music\\ 
 +C6: Learning to superoptimize programs\\ 
 +C7: Regularizing CNNs with Locally Constrained Decorrelations\\ 
 +C8: Generative Multi-Adversarial Networks\\ 
 +C9: Visualizing Deep Neural Network Decisions: Prediction Difference Analysis\\ 
 +C10: FractalNet: Ultra-Deep Neural Networks without Residuals\\ 
 +C11: Faster CNNs with Direct Sparse Convolutions and Guided Pruning\\ 
 +C12: FILTER SHAPING FOR CONVOLUTIONAL NEURAL NETWORKS\\ 
 +C13: The Neural Noisy Channel\\ 
 +C14: Automatic Rule Extraction from Long Short Term Memory Networks\\ 
 +C15: Adversarially Learned Inference\\ 
 +C16: Deep Information Propagation\\ 
 +C17: Revisiting Classifier Two-Sample Tests\\ 
 +C18: Loss-aware Binarization of Deep Networks\\ 
 +C19: Energy-based Generative Adversarial Networks\\ 
 +C20: Central Moment Discrepancy (CMD) for Domain-Invariant Representation Learning\\ 
 +C21: Temporal Ensembling for Semi-Supervised Learning\\ 
 +C22: On Detecting Adversarial Perturbations\\ 
 +C23: Understanding deep learning requires rethinking generalization\\ 
 +C24: Adversarial Feature Learning\\ 
 +C25: Learning through Dialogue Interactions\\ 
 +C26: Learning to Compose Words into Sentences with Reinforcement Learning\\ 
 +C27: Batch Policy Gradient Methods for Improving Neural Conversation Models\\ 
 +C28: Tying Word Vectors and Word Classifiers:​ A Loss Framework for Language Modeling\\ 
 +C29: Geometry of Polysemy\\ 
 +C30: PGQ: Combining policy gradient and Q-learning\\ 
 +C31: Reinforcement Learning through Asynchronous Advantage Actor-Critic on a GPU\\ 
 +C32: Learning to Navigate in Complex Environments\\ 
 +C33: Learning and Policy Search in Stochastic Dynamical Systems with Bayesian Neural Networks\\ 
 + 
 +<​html><​div id='​wednesday_afternoon'></​div></​html>​ 
 +====Wednesday Afternoon (April 26th, 4:30pm to 6:​30pm)==== 
 +C1: Learning recurrent representations for hierarchical behavior modeling\\ 
 +C2: Predicting Medications from Diagnostic Codes with Recurrent Neural Networks\\ 
 +C3: Sparsely-Connected Neural Networks: Towards Efficient VLSI Implementation of Deep Neural Networks\\ 
 +C4: HolStep: A Machine Learning Dataset for Higher-order Logic Theorem Proving\\ 
 +C5: Learning Invariant Representations Of Planar Curves\\ 
 +C6: Entropy-SGD:​ Biasing Gradient Descent Into Wide Valleys\\ 
 +C7: Amortised MAP Inference for Image Super-resolution\\ 
 +C8: Inductive Bias of Deep Convolutional Networks through Pooling Geometry\\ 
 +C9: Neural Photo Editing with Introspective Adversarial Networks\\ 
 +C10: A Learned Representation For Artistic Style\\ 
 +C11: Learning ​to Remember Rare Events\\ 
 +C12: Optimization as a Model for Few-Shot Learning\\ 
 +C13: Support Regularized Sparse Coding and Its Fast Encoder\\ 
 +C14: Discrete Variational Autoencoders\\ 
 +C15: Training Compressed Fully-Connected Networks with a Density-Diversity Penalty\\ 
 +C16: Efficient Representation of Low-Dimensional Manifolds using Deep Networks\\ 
 +C17: Semi-Supervised Classification with Graph Convolutional Networks\\ 
 +C18: Understanding Neural Sparse Coding with Matrix Factorization\\ 
 +C19: Tighter bounds lead to improved classifiers\\ 
 +C20: Why Deep Neural Networks for Function Approximation?​\\ 
 +C21: Hierarchical Multiscale Recurrent Neural Networks\\ 
 +C22: Dropout with Expectation-linear Regularization\\ 
 +C23: HyperNetworks\\ 
 +C24: Hadamard Product for Low-rank Bilinear Pooling\\ 
 +C25: Adversarial Training Methods for Semi-Supervised Text Classification\\ 
 +C26: Fine-grained Analysis of Sentence Embeddings Using Auxiliary Prediction Tasks\\ 
 +C27: Pointer Sentinel Mixture Models\\ 
 +C28: Reasoning with Memory Augmented Neural Networks for Language Comprehension\\ 
 +C29: Dialogue Learning With Human-in-the-Loop\\ 
 +C30Zoneout: Regularizing RNNs by Randomly Preserving Hidden Activations\\ 
 +C31: Learning to Play in a Day: Faster Deep Reinforcement Learning by Optimality Tightening\\ 
 +C32: Learning Visual Servoing with Deep Features and Trust Region Fitted Q-Iteration\\ 
 +C33: An Actor-Critic Algorithm for Sequence Prediction\\