The Reward Positivity Tracks Positive Reward Prediction Errors From Feedback to Cues During Reinforcement Learning.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Temporal difference (TD) learning theory predicts that reward prediction errors (RPEs) shift temporally during learning.
• The reward positivity (RewP) is an electrophysiological signal linked to anterior midcingulate cortex sensitivity to positive RPEs.
• It's hypothesized that RewP should transfer from outcome feedback to predictive cues as learning progresses, but this remains largely untested.

Purpose of the Study:

• To investigate the temporal dynamics of the reward positivity (RewP) during reinforcement learning.
• To test the core prediction of TD learning theory regarding the temporal backpropagation of RPEs.
• To examine individual differences in learning rates and their relationship with RewP shifts.

Main Methods:

• Electroencephalography (EEG) was recorded from 73 healthy adults during an extended probabilistic selection task (PST).
• RewP amplitude was measured at cue and feedback during early and late learning phases.
• Participants were categorized into rapid and slow learners, and Q-learning models were used to estimate learning rates.

Main Results:

• Evidence of temporal backpropagation was observed: RewP shifted from feedback to predictive cues as learning progressed.
• Rapid learners exhibited more pronounced RewP shifts, correlating with higher learning rates.
• The RewP disappeared from feedback responses late in learning, emerging instead at cue presentation.

Conclusions:

• This study provides the first clear evidence for temporal backpropagation of the reward positivity, supporting TD learning theory.
• The findings validate RewP as a neural marker for positive RPEs.
• Understanding these neural dynamics is crucial for characterizing reinforcement learning and individual differences.