Reinforcement learning models and their neural correlates: An activation likelihood estimation meta-analysis.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Psychiatry

Background:

• Reinforcement learning (RL) models motivated behavior using prediction error and expected value signals.
• Mesostriatal pathways and prefrontal cortex regions are implicated in RL, but human brain mapping is needed.

Purpose of the Study:

• To create unbiased maps of human brain representations of RL signals.
• To investigate the neural correlates of prediction error and expected value using meta-analysis.

Main Methods:

• Meta-analysis of functional magnetic resonance imaging (fMRI) studies employing algorithmic RL models.
• Analysis across diverse experimental paradigms to identify consistent activation patterns.

Main Results:

• Ventral striatum, midbrain, and thalamus showed reward prediction error signals, aligning with animal studies.
• Frontal operculum/insula processed prediction errors, especially for social rewards; amygdala and caudate were involved in specific task types.
• A posterior ventromedial prefrontal cortex region correlated with expected value.

Conclusions:

• A reproducible neural motif for reinforcement learning exists within cortico-striatal loops.
• Methodological factors like model fitting and spatial smoothing influence activation pattern reproducibility.