Brain networks for confidence weighting and hierarchical inference during probabilistic learning

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Classical learning algorithms struggle with unpredictable environmental changes.
• Optimal learning requires weighting prior knowledge and new evidence by their reliability, a process termed confidence weighting.

Purpose of the Study:

• To investigate if the human brain employs confidence weighting for probabilistic learning.
• To identify the neural mechanisms underlying confidence-based statistical learning.

Main Methods:

• Functional magnetic resonance imaging (fMRI) and ideal-observer analysis were used.
• Participants learned transition probabilities in auditory or visual sequences while their brain activity and confidence were recorded.

Main Results:

• Subjective confidence reports aligned with ideal observer predictions, demonstrating Bayes-optimal inference.
• Separate brain regions tracked prediction likelihood and confidence, with integration in the right inferior frontal gyrus.
• The right inferior frontal gyrus showed neural signatures of hierarchical processing for uncertainty.

Conclusions:

• Confidence weighting is a fundamental component of probabilistic learning in the human brain.
• The right inferior frontal gyrus implements a confidence-based statistical learning algorithm for sequential data.