Subject Adversarial Neural Network on EEG-Based Emotion Recognition

We propose a deep learning framework called the subject adversarial neural network to classify emotions based on electroencephalogram. Many previous emotion recognition works perform badly on the unseen test set. It is mainly because electroencephalogram was collected from different subjects. Electroencephalogram collected from different subjects has different feature distributions. In this paper, we use a temporal convolutional network as a feature extractor in the subject adversarial neural network to learn the invariant features across different subjects. We evaluate the emotion recognition efficiency of our proposal on the open dataset. The results show that our proposal provided 79.45\(\%\) and 76.3\(\%\) in mean accuracies for two emotion recognization (valence and arousal).