Optimal decision-making requires the brain to properly represent and compute with sensory uncertainty and not just the best sensory estimates. The Bayesian brain hypothesis postulates that brains represent sensory uncertainty and perform probabilistic computations, utilizing statistical generative models of the world to arrive at a decision. Multiple theories on how cortical populations represent and perform probabilistic computations with sensory uncertainty have been proposed, but we are only beginning to see concrete experiments testing these theories.
Making decisions, in all their complexity ranging from “simple” perception to high-level cognition, is arguably the cornerstone of brain function. At almost every moment, animals engage in sophisticated decision-making, combining prior experiences and knowledge about the world with sensory information from different sources and over different time scales.
At Walker Lab, we develop and utilize deep learning-based methods to analyze large-scale multimodal experimental data, overcoming the limitations of more conventional analysis techniques in making sense of the complex datasets to guide ongoing and future experiments.
Combining theories of probabilistic computations in the brain with population recordings and behavioral experiments carefully designed to test theoretical predictions, we work in tight collaboration with animal experimental labs to design and conduct experiments that combine state-of-the-art population recording techniques with complex behavioral paradigms involving decision-making under uncertainty, and subsequently apply novel deep learning-based models and analyses to elucidate the sensory representation and mechanisms underlying probabilistic computation in the brain.