Neural representations of beliefs in a multi-dimensional inference task.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Adaptive behavior relies on updating probabilistic beliefs.
• The neural mechanisms for these computations in distributed brain circuits are not well understood.

Purpose of the Study:

• To investigate how distributed brain circuits implement probabilistic belief updating.
• To identify neural representations of beliefs and confidence during inference.

Main Methods:

• Recorded neural activity from over 1,400 neurons in six brain regions of monkeys.
• Used a multi-dimensional inference task involving trial-and-error learning.
• Modeled behavior using Bayesian updating of beliefs.

Main Results:

• Neural representations of stimuli, rewards, and latent beliefs were widespread across brain regions.
• Population neural activity changes mirrored Bayesian belief updating, integrating prior beliefs with new evidence.
• Identified confidence representations independent of specific beliefs with distinct temporal dynamics.

Conclusions:

• Probabilistic inference arises from coordinated dynamics across distributed brain systems.
• Different brain regions contribute flexibly to belief updating based on computational demands.
• Neural representations of beliefs and confidence are dynamically updated during learning and decision-making.