Multi Task Learning for Predicting Music Video Clip and Participant Identification from Brain Responses

We designed a CNN-based Multi-task (MTL) classification model to predict music-video clips and identify participants from EEG responses to affective audio-visual stimuli from the DEAP dataset. Leveraging the capabilities of MTL models, participants are identified more accurately and clips are predicted better than single-task techniques. Variable train-test splits do not affect participant prediction task but show depreciation in clip prediction accuracy. It is true for both multi-task and singular task models. These findings show that MTL models may capture complex brain patterns with improved accuracy, making it a reliable EEG-based categorization solution.