Regularization, Action, and Attractors in the Dynamical "Bayesian" Brain.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Bayesian Brain Theory

Background:

• The influential "Bayesian brain" hypothesis posits the brain infers hidden causes of sensory input.
• This framework has driven significant research in cognitive and neuroscience.

Purpose of the Study:

• To present a generalization of the Bayesian brain hypothesis.
• To propose that the brain's primary task is tracking the probabilistic structure of observations directly.
• To reframe prior expectations, regularization, and behavioral goals as attractors in a dynamical system.

Main Methods:

• Theoretical generalization of Bayesian inference models.
• Conceptual framework integrating prior experience, regularization, and goal-directed behavior.
• Dynamical systems perspective on neural computation.

Main Results:

• The brain's core function is tracking observation structure, not solely inferring hidden states.
• Prior experience provides essential regularization for noisy observations.
• Behavioral goals function as attractive forces, similar to prior expectations.

Conclusions:

• This perspective refines the Bayesian brain framework by focusing on observable structures.
• It offers a unified view of prior expectations, regularization, and goals as attractors.
• The proposed model may stimulate new empirical and theoretical research by highlighting novel parallels in brain function.