The Anatomy of Inference: Generative Models and Brain Structure

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The brain infers environmental causes from sensory data using generative models.
• Active inference frames perception and action as approximate Bayesian inference, explaining many brain functions.
• Recent work links inference processes to neurobiological substrates, necessitating anatomical considerations.

Purpose of the Study:

• To review and develop the anatomical aspects of active inference process theory.
• To argue that generative models constrain brain region connectivity.
• To demonstrate how generative models explain anatomical brain architectures.

Main Methods:

• Reviewing existing literature on active inference and brain connectivity.
• Proposing that neuronal populations require input from their Markov blanket for inference.
• Illustrating the proposed anatomical constraints in perception, planning, attention, and movement.

Main Results:

• The structure of generative models dictates how brain regions should connect.
• Neuronal populations representing beliefs must receive input from their Markov blanket populations.
• This framework accounts for anatomical brain architectures across multiple domains.

Conclusions:

• Generative models provide a basis for understanding brain anatomy and connectivity.
• Anatomical theories of inference facilitate testable empirical hypotheses.
• Neuroimaging, neuropsychological, and electrophysiological methods can validate these hypotheses.