The Effect Of Reward Prediction Error On Familiarity And Recollection

Recognition memory is a part of the episodic memory system and an important indicator to measure memory accuracy.It is generally believed that recognition memory involves two processing: familiarity and recollection.Familiarity is an individual’s sense of familiarity with an old item,but the inability to extract detailed information.Recollection is the ability to extract details related to old projects.There are differences in memory strength between these two types of processing,with familiarity reflecting quantitative characteristics of memory strength information,and recollection reflecting qualitative characteristics.Error-driven learning theory believes that the mismatch between actual results and prediction results(ie,Prediction error,PE)is the main driving factor for learning.Reward prediction error(RPE)is a subset of prediction error that induces a prediction error signal during the learning process,thereby changing the encoding of relevant information and promoting learning and memory.The reward prediction error can be divided into signed reward prediction error(SRPE)and unsigned reward prediction error(URPE).Previous studies have found that both types of prediction error can promote recognition memory,but the effects on familiarity and recollection are unclear.Additionally,the learning process of reward prediction error can be affected by the framing effect.To maximize benefits,individuals allocate cognitive resources based on motivational differences caused by gain and loss contexts,which in turn affect learning.However,it is unclear whether the memory performance of reward prediction error is affected by the framing effect.To solve the above-mentioned problems,this paper uses three studies(8experiments),including frame effect and gradient condition variables,to investigate the effects of each attribute(valence,magnitude,and probability)of reward prediction error on recognition memory.and its processing.At the same time,taking advantage of the high temporal resolution of ERPs,we explore the encoding and retrieval process of reward prediction error and provide a basis for the system to understand its internal process and processing mechanism.In Study 1(Experiment 1 and Experiment 2),by setting different gradient ranges,the effects of reward prediction error valence on recognition memory were investigated in the gain and loss context,respectively.The results show that the memory performance of the gain context is better than that of the loss context;the memory performance of the positive reward prediction error is better than the negative reward prediction error in the gain context.This indicates that the reward prediction error valence only affects recognition memory in gain context,and has no effect on familiarity and recollection.In addition,the gradient range moderates the memory performance of the prediction error magnitude of the signed reward prediction error in the gain context.Study 2 examined the effect of reward prediction error magnitude on recognition memory by comparing different gradient conditions(Experiments 3 and 4)and framing effects(Experiments 5 and 6).Experiment 3 found that signed reward prediction error affected familiarity in the gain context while promoting recollection in the loss context.Unsigned reward prediction error promotes both familiarity and recollection in the loss context.This indicates that both signed and unsigned reward prediction errors had significant effects on recognition memory,but their memory strengths differed among different contexts.Experiment 4(compared to Experiment 3)expanded the reward gradient interval and found that signed reward prediction error promotes recollection in the gain and loss context.Unsigned reward prediction error promotes recollection in loss context.Combining the results of experiments 3 and 4,we found that the large gradient condition(than the small gradient condition)induced a stronger prediction error signal,indicating that the memory retrieval of the individual under the large gradient condition requires more sufficient detailed information,that is,more dependent recollection processing.To reveal the processing stages of reward prediction error in learning and retrieval.Experiments 5 and 6 used ERPs technology to examine the effect of reward prediction error on familiarity and recollection in the gain and loss context,respectively.The behavioral results found that in the gain context,reward prediction error magnitude is significantly different in response time during retrieval,while in the loss context,the recognition accuracy is significantly affected and depends on recollection processing.On the ERP results,in the learning stage,the magnitude of the reward magnitude error in both gain and loss context induced the FRN and P300 components,the former was related to the salience prediction error,and the latter was related to the outcome evaluation and context update.In the recognition test stage,in the gain context,the reward prediction error magnitude induced the FN400 component in the "remember" response,indicating that the background detail extraction of the stimulus related to the reward prediction error magnitude affected familiarity and was reflected in the FN400 component.In the loss context,reward prediction error magnitude inducted FN400 and LPC components in the "know" response,the former located in the frontal and parietal regions and the latter in the mid-parietal region.This indicates that familiarity and recollection processing are involved in the subject extraction of relevant materials and are reflected in the FN400 and LPC components respectively.The third study(Experiment 7 and Experiment 8)mainly investigated the effect of reward prediction error probability on recognition memory.Experiment 7 found that signed reward prediction error probability affects familiarity and recollection,while unsigned reward prediction error affects familiarity only.Experiment 8(compared to Experiment 7)narrowed down the probability range and used ERPs technology to further investigate its neural mechanism.The behavioral results found that there were significant differences in response times for the reward prediction error probability during retrieval;in the ERP results,in the learning stage,the reward prediction error probability induced FRN and P300 components in the mid-parietal and parietal regions;in the recognition test stage,reward prediction error probability induces FN400 components in "remember" responses.It shows that the familiarity processing will be affected when the subjects extract the background details of the relevant materials.Based on the above series of research results,this study mainly draws the following conclusions:(1)All attributes of reward prediction error can promote recognition memory but have different effects on familiarity and recollection;(2)Reward prediction error valence can promote recognition memory in gain context,but has no effect on familiarity and recollection;(3)Signed and unsigned reward prediction error can promote recognition memory,but the effects on familiarity and recollection are moderated by framing effects and gradient conditions;(4)There are differences in the neural mechanisms of reward prediction error in the learning and retrieval process under the gain and loss context;(5)Both signed and unsigned reward prediction error probability affects recognition memory,but the performance of familiarity and recollection is different.To sum up,the existing reinforcement learning theory and salience theory have failed to accurately explain the effect of reward prediction error on memory.That is,the memory performance under different reward prediction error conditions is not fixed,but will be affected by the properties of reward prediction error,frame effect,and gradient conditions,which is a dynamic process.