Abstract
In this study, we implement a Wasserstein GAN with Gradient Penalty (WGAN-GP)to generate multi-channel resting-state EEG data and assess the quality of thesynthesized signals through both visual and feature-based evaluations. Ourresults indicate that the model effectively captures the statistical andspectral characteristics of real EEG data, although challenges remain inreplicating high-frequency oscillations in the frontal region. Additionally, wedemonstrate that the Critic's learned representations can be reused for genderclassification task, achieving an out-of-sample accuracy, significantly betterthan a shuffled-label baseline and a model trained directly on EEG data. Thesefindings suggest that generative models can serve not only as EEG datagenerators but also as unsupervised feature extractors, reducing the need formanual feature engineering. This study highlights the potential of GAN-basedunsupervised learning for EEG analysis, suggesting avenues for moredata-efficient deep learning applications in neuroscience.