Graph Signal Representation of EEG for Graph Convolutional Neural Network

In this paper, we present an approach for graph signal representation of EEG toward deep learning-based modeling. In order to overcome the low dimensionality and spatial resolution of EEG, our approach divides the EEG signal into multiple frequency bands, builds an intra-band graph for each of them, and merges them with inter-band connectivity to obtain rich graph representation. The signal features on the vertices are also obtained from EEG. Finally, the graph signals are learned with graph convolutional neural networks. Experimental results on visual content identification using EEG are presented and various ways of defining intra-band and inter-band connections are examined.