UC San Diego

We present the results using single-trial analyses and pattern classifier to analyze Electroencephalography (EEG) data recorded during recognition memory experiments.

In Chapter 2, the details of the recorded data and the experimental paradigms for both location source information and frame color source information experiments were given, and the data recorded would be explored throughout the entire study.

In Chapter 3, we used subject-dependent leave-one-trial-out (LOTO) pattern classifiers to extract features related to recognition memory retrieval from the spatio-temporal information in single-trial EEG data during attempted memory retrieval. The results showed that location may be bound more tightly with the item than an extrinsic color association. The multivariate classification approach also showed that trial-by-trial variation in EEG corresponding to these ERP components were predictive of subjects’ behavioral responses.

In Chapter 4, we performed EEG classification in memory retrieval predictions using classifiers trained on a leave-one-subject-out (LOSO) cross-validation basis. We also compared the performance to that of classification using LOTO when trained on data for an individual subject. Unlike traditional single-trial EEG analysis performed within an individual subject, we show that it is possible to perform single-trial EEG classification using classifiers trained on different subjects.

In Chapter 5, we explored the separation of decision confidence and familiarity components in EEG data from recognition memory experiments. We first developed and tested a classifier designed to classify decision confidence on new trials. We then used this classifier to control for confidence in the selection of trials of familiarity and correct rejection. This allowedus to reveal a familiarity component that is of similar magnitude for recollection and familiarity judgements. This familiarity component revealed more of a frontal extent than obtained without confidence matching.

While confidence is often considered to be indexed by memory strength, in Chapter 6, we firstly showed that the confidence classifier trained with new item decision could be disassociate with the memory strength. Moreover, projecting old decisions onto the confidence classifier revealed the same confidence feature for both the old and new item decision and that the latepositive component (LPC) could be related to both confidence in old and new responses. By using linear regression models, we also showed that the difference between remember and know judgments in EEG recordings can be expressed by the difference in source memory, item memory and confidence.

In Chapter 7, we reviewed the single- and dual-process model for memory retrieval and showed how they both failed to explain the results from our trained CR-SN vs. CR-MN classifier. We then introduced our proposed model based on source memory, item memory, and confidence.