GANSER: A Self-supervised Data Augmentation Framework for EEG-based Emotion Recognition

Electroencephalography (EEG)-based affective computing has a scarcity problem. As result, it is difficult to build effective, highly accurate and stable models using machine learning algorithms, especially deep models. Data augmentation recently shown performance improvements in with increased accuracy, stability reduced overfitting. In this paper, we propose novel data framework, named the generative adversarial network-based self-supervised (GANSER). first combine training for EEG-based emotion recognition, proposed framework generates high-quality high-diversity simulated EEG samples. particular, utilize learn an generator force generated signals approximate distribution of real samples, ensuring quality augmented A transformation operation employed mask parts synthesize potential based on unmasked produce wide variety masking possibility during introduced as prior knowledge generalize classifier sample space. Finally, numerous experiments demonstrate that our method can improve recognition increase achieve state-of-the-art results.