Dorsal-Ventral Reinforcement Learning Network Connectivity and Incentive-Driven Changes in Exploration.

Area of Science:

• Neuroscience
• Cognitive Science
• Decision Science

Background:

• Probabilistic reinforcement learning (RL) involves decisions under uncertainty, influenced by internal models (model-based) or direct experience (model-free).
• Individual differences in balancing model-based and model-free control are affected by incentive motivation, but the impact of variable rewards is understudied.
• Neural and cognitive factors moderating reward effects on RL control strategies are largely unknown.

Purpose of the Study:

• To investigate how variable reward incentives influence the arbitration between model-based and model-free learning in decision-making.
• To identify individual differences in cognitive traits and neural signatures that moderate the effect of reward on RL control.
• To explore the relationship between functional brain connectivity and reward-modulated exploration-exploitation trade-offs.

Main Methods:

• A two-stage decision-making task with varying reward incentives was employed.
• Computational modeling, neuropsychological tests, and neuroimaging (resting-state functional connectivity) were utilized.
• Analyses were conducted across two independent datasets including both sexes.

Main Results:

• Increased reward prospect decreased exploration and shifted the balance towards model-based learning, consistent across datasets.
• Processing speed and analytical thinking style modulated the influence of reward on model-based/model-free control.
• Reduced exploration under high incentives correlated with increased cross-network coupling between ventral (stimulus valuation) and dorsal (action valuation) RL circuitry.

Conclusions:

• Reward prospect significantly alters the balance between exploration and exploitation, favoring model-based strategies.
• Cognitive traits and resting-state functional connectivity within RL networks are key moderators of reward's impact on decision-making.
• The integrity of connections between ventral and dorsal RL networks is associated with adaptive explore-exploit adjustments under changing incentives.