EEG Data Augmentation for Emotion Recognition with a Task-Driven GAN

The high cost of acquiring training data in the field emotion recognition based on electroencephalogram (EEG) is a problem, making it difficult to establish high-precision model from EEG signals for tasks. Given outstanding performance generative adversarial networks (GANs) augmentation recent years, this paper proposes task-driven method CWGAN generate high-quality artificial data. generated are represented as multi-channel differential entropy feature maps, and task network (emotion classifier) introduced guide generator during training. evaluation results show that proposed can with clearer classifications distributions more similar real data, resulting obvious improvements EEG-based