Temporal dynamics of prediction error processing during reward-based decision making.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Adaptive decision-making requires accurate reward representations.
• Reinforcement learning (RL) uses prediction error (PE) to update reward expectations.
• Previous EEG studies noted feedback-related potentials in performance monitoring but lacked temporal detail.

Purpose of the Study:

• To investigate the temporal sequence of feedback processing in reward-based decision-making.
• To clarify the specific role of feedback-related potentials (FRPs) in learning reward contingencies.
• To test the hypothesis that feedback processing involves sequential valence evaluation and PE magnitude representation.

Main Methods:

• Model-based single-trial analysis of EEG data.
• Utilized a reversal learning task.
• Employed prediction errors derived from a reinforcement learning model.

Main Results:

• Feedback outcomes are evaluated categorically by valence around 220ms post-feedback.
• Quantitative representation of prediction error (PE) magnitude emerges around 300ms, in parallel with valence evaluation.
• Feedback-related potentials reflect quantitative PE information crucial for learning, not just error awareness.

Conclusions:

• Brain processing of feedback is temporally organized, starting with valence and followed by PE magnitude.
• This sequential processing provides comprehensive information for updating reward expectations and guiding adaptive decisions.
• Feedback-related potentials are key neural signals carrying quantitative learning information in RL.