Search Papers | Poster Sessions | All Posters
Poster C123 in Poster Session C - Friday, August 9, 2024, 11:15 am – 1:15 pm, Johnson Ice Rink
Bayesian mechanics of learning in the brain
Chang Sub Kim1 (); 1Chonnam National University
The brain is a biological system orchestrating its embodied agent's perception, learning, and behavior in the environment. Recently, we elaborated on the brain theory of higher-order functions in a physics-guided manner. Our study revealed that the brain's Bayesian inference facilitates local, recurrent, and unsupervised neural dynamics, completing the perception-action closed loop. In this work, we extend our effort to incorporate `learning’ into the formulation by accounting for learning as inference as well and derive the governing equations unified with perception and motor behavior. Subsequently, using a parsimonious generative model, we show how the brain integrates the Bayesian mechanics of learning subject to a time-dependent sensory stream. As a result, attractors are manifested to form in neural phase space and make dynamic transitions during the learning period.
Keywords: Perception, learning, and behavior Variational Bayesian inference Attractor dynamics Prediction-error representation