Special Talk

December 6, 2024
16:00-17:30 JST (BST + 8:00)

In-person: CiNet 1F Conference Room
Online: Please sign up here (by noon on Dec. 5)

Talk Title: How does the brain model the world?: towards theory- and data-driven model of probabilistic perception

Edgar Walker
Assistant Professor
Computational Neuroscience Center
Department of Physiology and Biophysics
University of Washington

Host: Satoshi Nishida

Abstract:
Understanding how the brain transforms sensory information into decisions and behavior remains a fundamental challenge in computational neuroscience. This process involves two key operations: sensory encoding, where cortical populations represent information about the environment, and decoding, where downstream circuits extract and utilize this information for decision-making. Recent advances in large-scale neural recordings and deep learning have yielded powerful “encoding” models that predict neural population responses to complex natural stimuli with unprecedented accuracy. However, these models function largely as black boxes, offering limited insight into how the brain actually interprets sensory information using its internal model of the world to guide behavior.

Classical theories propose that the brain implements a generative model of the world, performing Bayesian inference during sensory perception. While this framework has been theoretically influential, empirical validation has largely been qualitative. In this talk, I will present a novel approach that bridges purely data-driven encoding models and theory-based approaches through deep learning architectures constrained by probabilistic inference principles. Unlike conventional deep learning models that solely predict neural responses, our theory-guided models may be used to uncover the brain’s generative model of the world, revealing interpretable components that link encoding and decoding operations. Finally, I will propose targeted experimental paradigms designed to test such theoretically-constrained models, opening new avenues for investigating how the brain implements probabilistic computation during perception and decision-making.