Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Spatial-Temporal Transformer with Curriculum Learning for EEG-Based Emotion Recognition

Created by
  • Haebom

Author

Xuetao Lin (Beihang University, Beijing, China, SKLCCSE, Beijing, China), Tianhao Peng (Beihang University, Beijing, China, SKLCCSE, Beijing, China), Peihong Dai (Beihang University, Beijing, China, SKLCCSE, Beijing, China), Yu Liang (Beijing University of Technology, Beijing, China), Wenjun Wu (Beihang University, Beijing, China, SKLCCSE, Beijing, China) (China)

Outline

This paper proposes a novel framework, SST-CL, to address two key challenges in EEG-based emotion recognition in brain-computer interface (BCI) systems: effectively integrating anomalous spatiotemporal neural patterns and robustly adapting to dynamic emotional intensity changes in real-world environments. SST-CL integrates a spatiotemporal transformer with curriculum learning to simultaneously extract spatiotemporal correlations and multiscale dependencies. A spatial encoder models inter-channel relationships, while a temporal encoder captures multiscale dependencies via a window-based attention mechanism. Furthermore, an intensity-aware curriculum learning strategy progressively guides learning from high-intensity to low-intensity emotional states. Experimental results on three benchmark datasets demonstrate state-of-the-art performance across a range of emotional intensity levels, and ablation studies validate the necessity of the architectural components and curriculum learning mechanism.

Takeaways, Limitations

•
Takeaways:
◦
A novel approach integrating spatiotemporal transformers and curriculum learning in EEG-based emotion recognition is presented.
◦
Effectively addressing abnormal spatiotemporal neural patterns and dynamic emotional intensity changes.
◦
Achieving cutting-edge performance across a range of emotional intensity levels.
◦
Verification of the effectiveness of the proposed architecture and curriculum learning.
•
Limitations:
◦
Further research is needed to evaluate the generalization performance of the proposed method (testing on various datasets and emotion types).
◦
Real-time emotion recognition performance evaluation in real-world environments is needed.
◦
Further research is needed to determine optimal parameters for curriculum learning strategies.
◦
Possible dependency on specific EEG equipment or datasets.
👍
Made with Slashpage