Towards EEG Generation Using GANs for BCI Applications

Abstract

Brain-computer interface has been always facing serious data-related problems such as lack of the sufficient data and data corruption. Artificial data generation is a potential solution to address these issues. Among generative techniques, the method of generative adversarial networks (GANs) with the successful applications in image processing has gained a lot of attention. The application of GANs for time-series data generation is a recent growing topic that first of all its feasibility needs to be assessed. In the present study, we investigate the performance of GANs in generating artificial electroencephalogram (EEG) signals. The results suggest that the generated EEG signals by GANs resemble the temporal, spectral, and spatial characteristics of real EEG. It thus opens new perspectives for further research in this area.