EEG Emotion Recognition With Uncertainty-Aware Contrastive Learning and Frequency-Aware Self-Attention

Area of Science:

• Neuroscience
• Machine Learning
• Human-Computer Interaction

Background:

• Electroencephalography (EEG) emotion recognition is crucial for advancing human-machine interaction.
• Existing algorithms face challenges with unclear decision boundaries and noise in physiological signals.
• Robust EEG-based emotion recognition requires addressing signal noise and improving classification accuracy.

Purpose of the Study:

• To develop a novel framework, UACL-Net, for enhanced EEG emotion recognition.
• To address limitations of unclear decision boundaries and signal noise in current methods.
• To improve the robustness and accuracy of emotion recognition from EEG data.

Main Methods:

• Developed UACL-Net, integrating Uncertainty-Aware Contrastive Learning (UACL) and Frequency-Aware Self-Attention (FASA).
• UACL employs a multivariate Gaussian distribution to define the latent space, enhancing inter-class separation.
• FASA utilizes self-attention on frequency-domain components to adaptively reduce noise and capture temporal dependencies.

Main Results:

• Achieved high accuracy across four benchmark datasets: SEED (94.88%), DEAP (98.71%), DREAMER (96.91%), and FACED (99.29%).
• Demonstrated significant improvements in robustness and accuracy compared to state-of-the-art methods.
• Validated the effectiveness of UACL for clearer decision boundaries and FASA for noise reduction and dependency capture.

Conclusions:

• UACL-Net provides a robust and effective solution for EEG emotion recognition.
• The proposed framework significantly advances the field by overcoming key challenges in EEG signal processing.
• This work offers a promising direction for more sophisticated and reliable human-machine interactions.