WeC03 - Neurocomputation: Learning, Dynamics, and Optimization

Recent advances at the intersection of control theory and neuroscience have revealed novel mechanisms by which dynamical systems perform computation. These advances encompass a wide range of conceptual and computational ideas, used for model learning and training, memory retrieval, data-driven control, and optimization. This tutorial will highlight neuro-inspired approaches to computation that aim to improve scalability, robustness, and energy efficiency across such tasks, bridging the gap between artificial and natural systems.

Particular emphasis will be placed on energy-based and equilibrium models that encode information through gradient flows and attractor landscapes. By connecting these theoretical frameworks to their physical implementations in neuromorphic systems, the tutorial aims to steer the discussion beyond conventional feedforward and backpropagation-based approaches in artificial intelligence (AI).

The program will build a coherent trajectory from foundational to emerging models of neurocomputation. It will introduce classical formulations that demonstrate how dynamical models learn and compute through evolution toward equilibrium. These models include Hopfield networks, continuous-time recurrent neural networks, and control-theoretic formulations such as state-space learning. From this foundation, we will extend to modern developments that generalize these classical models: dense associative memory models for high-capacity storage, oscillator- and wave-based recurrent neural networks for large scale optimization, and proximal-descent models for composite optimization and constrained learning. We expect participants to learn how control-theoretic tools can guide the development of the next generation of neuromorphic computing systems.

Schedule of Talks