Prediction of Subsequent Memory Effects Using Convolutional Neural Network

Abstract

Listen

Translate article

Differences in brain activity have been associated with behavioral performance in memory tasks; in order to understand memory processes, previous studies have explored them and attempted to predict when items are later remembered or forgotten. However, reported prediction accuracies are low. The aim of this research is to predict subsequent memory effects using a convolutional neural network. We additionally compare different methods of feature extraction to understand relevant information related to memory processes during pre and on-going stimulus intervals. Subjects performed a declarative memory task while electroencephalogram signals were recorded from their scalp. The signals were epoched regarding stimulus onset into pre and on-going stimulus and used for prediction evaluation. A high prediction accuracy was obtained when using convolutional neural networks (pre-stimulus: 71.64% and on-going stimulus: 70.50%). This finding showed that it is possible to predict successful remember items on a memory task using a convolutional neural network, with a higher accuracy than conventional methods.