Dynamics underlying learning in complex environments

ongoing

Learning and adaptation play a central role in interacting with a changing environment, yet the neural substrate underlying such flexible, goal-directed behaviors has remained elusive. Neuroscience experiments have traditionally focused on how individual brain regions perform abstract, highly simplified tasks. However, recent technological advances, coupled with powerful AI-accelerated software, have rapidly enabled the monitoring of large populations of neurons over many days, across multiple brain regions, and during increasingly complex, naturalistic behaviors. Yet even with such data within our reach, we still lack the theoretical and quantitative tools that are necessary to understand the fundamental principles guiding learning in neural populations. DULCE aims to fill this gap by establishing a unified framework to understand learning in complex environments. The core hypothesis of DULCE is that in naturalistic conditions, learning engages multiple co-occurring learning processes that are distributed across the brain, and which work together to reshape neural dynamics to perform new tasks. As such, DULCE aims to uncover the behavioral, population- level, and synaptic learning rules responsible for guiding learning in complex environments. By interweaving statistical modelling, dynamical systems theory and machine learning, DULCE will: i) Develop hierarchical models of behavior that can disentangle the rules governing simultaneously occurring learning processes. ii) Provide a unified theory of how region-specific learning rules in the cortex, cerebellum, and striatum coordinate to form a distributed learning system. iii) Develop interpretable dimensionality reduction methods to identify the rules governing how task-relevant dynamics evolve in large-scale neural data over learning. Through this three-pronged attack, DULCE aims to lay the foundation necessary to uncover the neural mechanisms controlling the Dynamics Underlying Learning in Complex Environments.