Data-driven detection of memory encoding from EEG in an audiovisual task

Reliably detecting memory encoding can help diagnose memory impairments and provide insights into neural correlates of memory formation from electroencephalogram (EEG). However, there is little prior literature that can be used to drive feature extraction to detect encoding in audiovisual memory tasks. Our objective is to explore the use of an explainable neural network to detect memory encoding from EEG while participants watch videos. We use EEG-ITNet in single- and multi-user scenarios and show that detecting successful memory encoding from EEG data using audiovisual stimuli is possible, particularly when recordings from multiple participants are used to train the models. In the future we will explore the features identified by EEG-ITNet to detect a neural signature of memory encoding. Our framework can also be utilized for data-driven feature discovery in tasks where there is little prior knowledge.