A computational model of canonical cortical microcircuits for dynamic Bayesian inference and control as inference

Area of Science:

• Computational neuroscience
• Systems neuroscience
• Cognitive neuroscience

Background:

• Canonical cortical microcircuits (CCMs) form a conserved six-layer structure in the mammalian neocortex, vital for cognitive functions.
• Current computational models often lack detailed cell-type and connectivity representations, limiting understanding of CCMs' mechanisms.

Purpose of the Study:

• To develop a biologically plausible computational model of CCMs for dynamic Bayesian inference (DBI) and control as inference (CAI).
• To elucidate the computational mechanisms underlying cognitive functions mediated by CCMs.

Main Methods:

• Applied the structure-constrained interface decomposition (SCID) method to build CCM models.
• Explicitly assigned computational roles to neuronal populations and incorporated inhibitory neuron contributions.
• Simulated a mouse lever-push/pull task for perceptual decision-making and control.

Main Results:

• The model successfully reproduced key behavioral features, including psychometric curves and "change of mind" behaviors.
• Layer-specific perturbation simulations yielded experimentally testable predictions about cortical layer functions.
• Demonstrated a unified account of CCM functions in sensory and motor cortices based on inference and control duality.

Conclusions:

• The developed CCM model provides a unified framework for understanding sensory inference and motor control.
• The study highlights the functional roles of different cortical layers and offers testable predictions for future research.