Brain systems for probabilistic and dynamic prediction: computational specificity and integration.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience

Background:

• Functional specialization in the brain is often defined by sensory or behavioral domains.
• An alternative view proposes that brain systems are specialized by the computational operations they perform.

Purpose of the Study:

• To investigate whether distinct computational strategies (reinforcement-learning vs. dynamic forward modeling) recruit separable neural systems.
• To dissociate brain activity based on computational function, rather than sensory or behavioral domain, within a single task.

Main Methods:

• Functional magnetic resonance imaging (fMRI) was used to monitor brain activity.
• Participants performed a spatial prediction task where the precision of two distinct predictive models was manipulated.
• This manipulation induced shifts in computational strategies between reinforcement-learning and dynamic forward modeling.

Main Results:

• Activity in brain systems typically associated with reward learning and motor control was dissociated based on the computational strategy employed.
• This dissociation occurred even when both systems were engaged in making parallel predictions for the same event.
• A parietal cortex region showed sensitivity to prediction divergence and is proposed to integrate the two predictive modes.

Conclusions:

• Brain systems can be functionally specialized by their computational roles, independent of sensory or behavioral domains.
• Distinct computational strategies, even within the same task, engage separable neural networks.
• Parietal cortex plays a role in integrating information from different predictive computational models for unified behavioral output.