Neural variability in the default mode network compresses with increasing belief precision during Bayesian inference.

Area of Science:

• Neuroscience
• Cognitive Science
• Decision Making

Background:

• Intelligent agents require learning latent environmental states from observations for optimal decision-making.
• Bayesian frameworks suggest evidence integration over time refines state beliefs by reducing uncertainty.
• The neural mechanisms underlying increased belief precision during learning remain unclear.

Purpose of the Study:

• To investigate how increasing belief precision during learning is reflected in neural activity.
• To test the hypothesis that temporal neural variability scales with uncertainty reduction during learning.
• To explore the relationship between neural variability, learning accuracy, and prior beliefs.

Main Methods:

• Functional magnetic resonance imaging (fMRI) to measure blood-oxygen-level-dependent (BOLD) signal variability (SDBOLD) in 47 healthy adults.
• Analysis of SDBOLD across successive trials of a learning task involving decision-related evidence.
• Computational modeling of participant behavior to assess prior beliefs and uncertainty reduction.

Main Results:

• BOLD signal variability (SDBOLD) compressed with increased exposure to decision-related evidence.
• Greater SDBOLD compression, particularly in default mode network regions, was observed in more accurate participants.
• Computational modeling indicated that accurate learners had flatter prior beliefs, enabling greater uncertainty reduction reflected in SDBOLD changes.

Conclusions:

• Neural variability compresses as belief precision increases during effective learning.
• This compression, especially in default mode network regions, may reflect the brain's refinement of latent state beliefs.
• Findings propose a neural mechanism for learning the probabilistic nature of the environment through flexible uncertainty reduction.