Paper Title
Enhanced Multimodal Emotion Recognition Using Deep Representation Learning and Physiological Signal Augmentation

Abstract
Obstacles to emotion detection with physiological signals, especially EEG, include small sample sizes and signal quality fluctuation. A unique strategy that combines contrastive learning, GNNs, and (GANs) is put forth to overcome these problems. By increasing dataset quantity and variety, GAN-based data augmentation strengthens the resilience of the model. By lessening the detrimental effects of intra- and inter-subject vari- ability in EEG data, contrastive learning ensures that emotions are represented more accurately. To classify emotions, a GNN is used to learn the connections between the features that have been extracted. Experimental results show better accuracy in emotion categorization for both valence and arousal dimensions, with greater performance on the DEAP and MAHNOB-HCI datasets. The separate contributions of GAN, contrastive learning, and GNN to the overall model performance are shown via ablation research. The findings demonstrate how deep learning methods may improve EEG-based emotion identification and solve issues with cross-subject variability. Keywords - EEG, Emotion Detection, Contrastive Learning, Graph Neural Networks, GAN-Based Augmentation