Generative Replay with Cross-attention Graph Transformer for Continual Brain EEG-based Emotional Tracking in HCI

Human-computer interaction (HCI) systems increasingly have been using real-time brain electroencephalogram (EEG)-based emotion tracking to provide a better user experience. Although catastrophic forgetting and inefficient feature extraction still pose challenges in EEG-based models, continual learning remains a significant challenge. To address these issues, we propose a Generative Replay with the Cross-Attention Graph Transformer (GR-CAGT) model, which integrates past knowledge with learning new tasks. We also consider our approach in applying to two benchmark datasets, the Dataset for Emotion Analysis using Physiological (DEAP) and MAHNOB-HCI. In our methodology, generative replay is utilized for memory retention, a graph neural network (GNN) is applied for learning structured EEG features, and a cross-attention transformer is employed to capture temporal dependencies. The results of experiments, which exceed 99.23% on DEAP and 99.15% on MAHNOB-HCI, outperform traditional models. We confirm the efficacy of our approach through ablation studies, where we demonstrate that performance drops when key components are removed (generative replay: 86.42% or EEG normalization: 88.76%). Our model is state-of-the-art in terms of both accuracy and robustness, and it also adapts well to continual learning situations. This demonstrates the value of the attention mechanism and generative replay for tackling catastrophic forgetting. The pattern of EEG signal amplitude is further explored as a potential area for future research efforts in adaptive learning strategies and multimodal integration to enhance emotion-tracking capabilities for human-computer interaction (HCI) applications.