Jason MacLean (UChicago): Schmidt AI in Science Speaker Series

Organized by the University of Chicago’s Eric and Wendy Schmidt AI in Science Fellowship Program.

Agenda
4:00pm – 4:45pm:  Presentation
4:45pm – 5:00pm:  Q&A
5:00pm – 5:30pm: Reception

NEW MEETING LOCATION
Data Science Institute
5460 S University Ave
Room 105

Title: Computing with Spikes: Dynamic Excitation–Inhibition Balance in Biological and Artificial Networks

Abstract: Neocortical computation emerges from the dynamic interplay of excitation and inhibition within sparsely connected, heterogeneous recurrent networks. Information is not represented solely by average firing rates but by the precise timing and coordination of spikes. Using two complementary approaches, examining a canonical computation in mouse visual cortex and training spiking neural network (SNN) models constrained by the statistical features of cortical connectivity and spiking dynamics, we identify shared dynamical principles by which inhibition enables spike-based computation. In vivo, divisive normalization, a canonical cortical operation, reflects transient modulation of inhibitory neuron activity driven by the temporal structure of subcortical input. Across neurons, normalization strength varies systematically with heterogeneity in excitatory–inhibitory balance and moment-to-moment coupling within local populations, indicating that computation arises from dynamic coordination within recurrent networks. In parallel, task-trained SNNs, push–pull inhibitory motifs emerge with training and maintain opponent balance between excitatory and inhibitory units. This structured inhibition enables precise spike-time coordination and stable computation. Disrupting spike timing by only a few milliseconds, or relaxing biological realism, significantly impairs task performance, underscoring the causal role of temporal precision and dynamically evolving inhibition in recurrent information processing. Together, these results demonstrate that the cortex computes through continuously evolving spike interactions that reconfigure excitatory–inhibitory coupling over time. This framework unifies canonical operations such as normalization with emergent circuit motifs in trained spiking networks, providing a mechanistic link between biological and artificial systems that compute with spikes.

Bio: Jason MacLean, Professor of Neurobiology at The University of Chicago. MacLean’s area of expertise is computational neuroscience. The primary focus of our group is to apply a diversity of analytical tools, including network science approaches to corticalcircuit dynamics in order to: 1) establish the higher-order cellular and synaptic mechanisms that propagate spikes, 2) to build improved encoding and decoding models of single-trial circuit activity in behaving mammals, and 3) to compare and contrast sensory and motor areas of neocortex. MacLean is also a member of the NSF-Simons National Institute for Theory and Mathematics in Biology (NITMB), where he is collaborating with fellow NITMB member and University of Chicago Associate Professor of Organismal Biology and Anatomy Stephanie Palmer on the NITMB internal research project ‘Quantifying Natural Movement Variation in the Brain and Behavior.’