UC Davis

Human electrophysiological scalp signals were initially recorded by Hans Berger in 1924. One hundred years later, we have seen an explosion of findings using the electroencephalogram (EEG) to examine neurocognitive processes, and the field continues to push forward with increased use of ever more complex methods. In this dissertation, we first acquaint ourselves with canonical event related potential (ERP) components that have been extensively studied via univariate methods and then discuss how we can leverage machine learning to better detect these components. In the second part of this work, we ask whether a canonical ERP component, the P3b, is related to enhanced working memory representations via behavior, univariate analyses, and decoding of the information content in the neural signals. In the final chapter we venture away from “ERP components” and use a decision-making paradigm and machine learning techniques to ask whether working memory is reactivated to facilitate credit assignment during contingency learning. We applied machine learning to both voltages and alpha band power to determine whether we can decode the sensory features of memories at feedback epochs when items are brought back to mind. This work highlights the effectiveness of machine learning methods in the analysis of noninvasive human electrophysiological data and explores the information content of working memory representations as assessed via neural signals.