Attention-Based CNN Capturing EEG Recording's Average Voltage and Local Change

Xiaodong Qu

Abstract: The attention mechanism is one of the most popular deep learning techniques in recent years and it is arguably able to produce human-interpretable results. In this research, we developed a classification model combining two self-attention modules and a convolutional neural network. This model achieved benchmark or superior performance on two electroencephalography (EEG) recording datasets. Moreover, we demonstrated that the self-attention modules were able to capture features, including average voltage of signal features and instant voltage change of the EEG signals, by visualizing the attention maps they produced.