Active inference leads to Bayesian neurophysiology.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Theoretical Biology

Background:

• The neuronal mechanisms underlying the free-energy principle and active inference remain incompletely understood.
• The free-energy principle posits that biological systems, including the brain, minimize surprise by inferring causes of sensory input.
• Previous work has linked brain function to Bayesian inference but lacked a comprehensive neuronal substrate explanation.

Purpose of the Study:

• To review and elucidate the potential neuronal substrates that implement the free-energy principle and active inference.
• To connect neural network dynamics and plasticity to Bayesian inference and variational free energy minimization.
• To explore how neural network structure can represent generative models and formalize prior beliefs.

Main Methods:

• Introduction to the foundational concepts of the free-energy principle.
• Mathematical formulation linking neural dynamics and plasticity to Bayesian inference and free-energy minimization.
• Discussion of empirical approaches to examine neuronal substrates of beliefs and inference.

Main Results:

• Neural network dynamics minimizing a cost function can be mathematically equated to Bayesian inference minimizing variational free energy.
• The free-energy principle provides a universal characterization of neural networks, with network structure representing generative models.
• Formal association of neural network properties with prior beliefs regulating inference and learning is enabled.

Conclusions:

• The free-energy principle offers a unifying framework for understanding brain function, psychology, and biological phenomena via Bayesian inference.
• This perspective provides a deeper understanding of neuronal mechanisms underlying basic psychology and psychiatric disorders.
• Future research should focus on empirically examining the proposed neuronal substrates for beliefs and inference.