Feedback-related Negativity Codes Prediction Error but Not Behavioral Adjustment during Probabilistic Reversal Learning

摘要

We assessed electrophysiological activity over the medialnfrontal cortex (MFC) during outcome-based behavioral adjustmentnusing a probabilistic reversal learning task. During recording,nparticipants were presented two abstract visual patterns onneach trial and had to select the stimulus rewarded on 80% ofntrials and to avoid the stimulus rewarded on 20% of trials. Thesencontingencies were reversed frequently during the experiment.nPrevious EEG work has revealed feedback-locked electrophysiologicalnresponses over the MFC (feedback-related negativity;nFRN), which correlate with the negative prediction errorn[Holroyd, C. B., & Coles, M. G. The neural basis of human errornprocessing: Reinforcement learning, dopamine, and the errorrelatednnegativity. Psychological Review, 109, 679–709, 2002]nand which predict outcome-based adjustment of decision valuesn[Cohen, M. X., & Ranganath, C. Reinforcement learning signalsnpredict future decisions. Journal of Neuroscience, 27, 371–378,n2007]. Unlike previous paradigms, our paradigm enabled us tondisentangle, on the one hand, mechanisms related to the rewardnprediction error, derived from reinforcement learningn(RL) modeling, and on the other hand, mechanisms related tonexplicit rule-based adjustment of actual behavior. Our resultsndemonstrate greater FRN amplitudes with greater RL modelderivednprediction errors. Conversely expected negative outcomesnthat preceded rule-based behavioral reversal were notnaccompanied by an FRN. This pattern contrasted remarkablynwith that of the P3 amplitude, which was significantly greaternfor expected negative outcomes that preceded rule-based behavioralnreversal than for unexpected negative outcomes thatndid not precede behavioral reversal. These data suggest thatnthe FRN reflects prediction error and associated RL-based adjustmentnof decision values, whereas the P3 reflects adjustment ofnbehavior on the basis of explicit rules. ■