Analog Computing for Associative Memory

This talk presents how continuous physical systems provide a framework for understanding associative memory, offering insight into stability, capacity, and interference, while also directly realising these computations in hardware. Extending classical Hopfield models, it highlights how higher-order interactions in continuous systems (e.g., Kuramoto networks) enhance stability, suppress interference, and significantly improve memory capacity. The presentation also introduces Attractor-Keyed Memory (AKM), where retrieval and routing are unified through system dynamics, pointing toward efficient and interpretable analog machine learning hardware.