Computational neuropsychology and active inference


Computational neuropsychology and active inference

Thomas Parr


Recent approaches to theoretical neurobiology assume that the brain possesses a generative model that is used to infer the causes of sensory data. This implies perception is a process of optimising posterior beliefs, while actions represent experiments to disambiguate between perceptual hypotheses. Under this view, neurological and psychiatric syndromes result from ‘broken’ generative models that represent a poor fit to a patient’s environment. In this talk, I will overview some of our recent work using active inference, as formulated for a Markov decision process, to examine the influence of prior beliefs on behaviour – with a focus on the active visual system. This enables a quantitative phenotyping of neuropsychological syndromes in which visuospatial exploration is disrupted. In brief, we can express computational lesions (for example, those that underwrite visual neglect) as alterations in the parameters of prior probability distributions. Doing so allows us to quantify the evidence associated with different plausible explanations for pathological behaviour.

About the author

CeMEAI administrator