Semi-supervised EEG emotion recognition with Discriminative Graph Transformer Model
Oral presentation
Research approach
Combined a two-layer GCN for local sample relationships with a Graph Transformer that incorporates sample similarity into attention. Classified emotions from a graph of EEG DE features with limited labels.
Results
Subject-wise transductive evaluation on SEED-IV. With 60, 80 and 100 labeled samples out of 2,505, accuracy was 83.28 ± 3.70%, 88.22 ± 4.48% and 93.83 ± 1.87%, respectively.
All publications