UNIFYING NEURAL POPULATION DYNAMICS, MANIFOLD GEOMETRY, AND CIRCUIT STRUCTURE

A central aim of neuroscience is to understand how the dynamics of neural circuits give rise to cognitive functions such as perception, attention, and decision-making. Cortical neural circuits are hierarchical and recurrent, resulting in rich temporal dynamics of individual neurons and distributed selectivity across the population. Classical neural circuit models capable of characterizing cognitive processes struggle to account for this complexity of cortical responses. Recent approaches leveraging heterogeneous neural networks address this complexity by characterizing activity in terms of interactions among latent states. In these lecture notes, we highlight recent work aimed at increasing the interpretability of these models and relating them to classical circuit models. These new analytical approaches connect neural population dynamics, geometry of latent manifolds, and the underlying circuit structure to enable mechanistic insights into cognitive processes.