Bayesian brain

The Bayesian brain is the hypothesis that neural computation is fundamentally probabilistic inference — that the brain maintains and updates probability distributions over states of the world, rather than computing point estimates or fixed representations. On this view, perception is not the reconstruction of a pre-given reality but the inversion of a generative model: the brain infers the most probable causes of sensory input given prior expectations and the likelihood of observations under different hypotheses.

The hypothesis connects to predictive coding and the free energy principle, both of which formalize neural computation as the minimization of prediction error or surprise. It also connects to statistical learning: the brain's priors are not innate in any fixed sense but are themselves learned through exposure to the statistical structure of the environment. The Bayesian brain is not a claim that neurons literally implement Bayes' theorem in floating-point arithmetic. It is a claim that the functional architecture of neural computation approximates Bayesian inference through distributed, dynamical mechanisms that are subject to the same resource constraints — energy, time, precision — that govern all biological systems. The Neural coding problem — how neural populations represent probability distributions — remains one of the most active research areas in computational neuroscience.