Active Inference and Belief Propagation in the Brain

Explore deep temporal models in the brain through this distinguished lecture by Karl Friston from University College London. Delve into mathematical accounts of active inference to understand how the brain simulates behavior and electrophysiological responses using hierarchical generative models of discrete state transitions. Learn how evidence accumulation occurs across distinct temporal scales, enabling inferences about narratives and temporal scenes through Bayesian belief updating and associated neuronal processes. Examine how these principles reproduce epistemic foraging behaviors observed in reading, including perisaccadic delay period activity and local field potentials. Discover how deep model structures simulate responses to both local violations (such as font type changes) and global violations (such as semantic inconsistencies), reproducing mismatch negativity and P300 responses respectively. Understand how first principles can constrain our comprehension of computational architectures in the brain and the functional imperatives that apply to neuronal networks, providing insights into the fundamental mechanisms underlying neural computation and active inference processes.