A theory of cortical responses

Area of Science:

• Neuroscience
• Computational Neuroscience
• Perception

Background:

• The brain's sensory systems evolved to infer the causes of sensory input.
• Perceptual inference and learning are complex processes not fully explained by traditional models.
• Helmholtz's ideas on perception can be reformulated using modern statistical theories.

Purpose of the Study:

• To present a unified theoretical framework for understanding evoked brain responses and their generation.
• To explain neurobiological facts using a statistical model of perceptual inference and learning.
• To connect anatomical, physiological, and psychophysical attributes of the brain within a single perspective.

Main Methods:

• Formulating Helmholtz's perception theories in terms of modern statistical inference.
• Applying the principle of minimizing free energy (from statistical physics) to perceptual inference and learning.
• Utilizing empirical Bayes and hierarchical models to represent how sensory input is caused.

Main Results:

• A single principle, minimizing free energy, explains both perceptual inference and learning.
• Cortical responses are interpreted as the brain's effort to minimize free energy and encode stimulus causes.
• The model predicts hierarchical cortical organization, reciprocal connections with functional asymmetry, associative and spike-timing-dependent plasticity.
• It accounts for receptive field effects, endogenous evoked responses, repetition suppression, mismatch negativity (MMN), and P300.

Conclusions:

• Minimizing free energy provides a biologically plausible mechanism for perception and learning.
• Hierarchical models enable dynamic and context-sensitive prior expectations.
• This unified framework explains a wide range of brain functions and phenomena, offering insights into cortical coupling.