Cortical-like dynamics in recurrent circuits optimized for sampling-based probabilistic inference.

Area of Science:

• Computational neuroscience
• Systems neuroscience
• Neural dynamics

Background:

• Sensory cortices exhibit complex dynamical features like noise and oscillations.
• These features lack a unified theoretical explanation.
• Understanding these dynamics is crucial for deciphering neural computation.

Purpose of the Study:

• To develop a unifying model for ubiquitous dynamical features in sensory cortices.
• To investigate the computational function underlying these neural dynamics.
• To bridge the gap between theoretical models and biological observations.

Main Methods:

• Trained a recurrent excitatory-inhibitory neural circuit model of a visual cortical hypercolumn.
• The model was optimized to perform sampling-based probabilistic inference.
• Analyzed model outputs and compared them with electrophysiological recordings from awake monkeys.

Main Results:

• The optimized network model exhibited key biological properties: divisive normalization, stimulus-modulated noise, inhibition-dominated transients, and gamma oscillations.
• These dynamical features were found to play functional roles in accelerating inference.
• Model predictions were validated through novel analyses of monkey electrophysiological data.

Conclusions:

• Basic motifs of cortical dynamics arise from the efficient implementation of fast sampling-based inference.
• The unifying model provides a principled account for diverse neural phenomena.
• The study predicts testable properties of neural dynamics in future experiments.