Ryan Smith: Active inference as a computational framework for modeling empirical data скачать в хорошем качестве

Ryan Smith: Active inference as a computational framework for modeling empirical data

CCNB Seminar Series is hosted by the Center for Cognitive Neuroscience Berlin. Twitter: @CCNBerlin Title: Cellular mechanisms of conscious processing Date: 09.05.2022 Guest: Ryan Smith Affiliation: Laureate Institute for Brain Research Abstract: The aim of this talk is to introduce the active inference framework and how it can be applied in empirical research within neuroscience and psychiatry. Active inference is an influential theory of perception, learning, and decision-making based on approximate Bayesian inference. While taking many possible forms, it is most often implemented as a partially observable Markov decision process (POMDP) in discrete time with discrete state and outcome spaces. Perception and learning in these models is accomplished through message passing schemes and minimization of variational (or marginal) free energy. Selection of action sequences (policies) is accomplished by minimizing the expected free energy of future observations - which motivates a trade-off between information-seeking and reward-seeking. I will provide a walkthrough of each of these elements of active inference, and how the framework can be used to simulate neural data for use in fMRI/EEG studies. I will also review previous applications of this framework within empirical studies of behavior in computational psychiatry.