Contrastive Learning of Subject-Invariant EEG Representations for Cross-Subject Emotion Recognition

EEG signals have been reported to be informative and reliable for emotion recognition in recent years. However, the inter-subject variability of emotion-related still poses a great challenge practical applications EEG-based recognition. Inspired by neuroscience studies on correlation, we proposed Contrastive Learning method Inter-Subject Alignment (CLISA) tackle cross-subject problem. learning was employed minimize differences maximizing similarity signal representations across subjects when they received same emotional stimuli contrast different ones. Specifically, convolutional neural network applied learn aligned spatiotemporal from time series contrastive learning. The were subsequently used extract differential entropy features classification. CLISA achieved state-of-the-art performance our THU-EP dataset with 80 publicly available SEED 15 subjects. It could generalize unseen or testing. Furthermore, learned provide insights into mechanisms human processing.