A Multi-Teacher Distilling Framework With Data Privacy for EEG Emotion Recognition.

Area of Science:

• Neuroscience
• Machine Learning
• Data Privacy

Background:

• Subject-independent electroencephalography (EEG) emotion recognition faces challenges with limited data, poor cross-domain transfer, and suboptimal feature extraction.
• Existing methods struggle to balance performance enhancement with data privacy concerns.

Purpose of the Study:

• To develop an innovative framework for subject-independent EEG emotion recognition that enhances performance while preserving data privacy.
• To address limitations in data availability, cross-domain knowledge transfer, and feature extraction.

Main Methods:

• A novel multi-teacher knowledge distillation framework with data privacy is proposed, utilizing sequentially trained subnets without data sharing.
• A multi-teacher knowledge distillation strategy with knowledge filters and adaptive losses enhances cross-domain transfer.
• A spatio-temporal gate module is introduced for efficient feature extraction and channel selection.

Main Results:

• The proposed method achieved a 2% performance improvement on the DEAP dataset compared to state-of-the-art approaches.
• Experimental results demonstrate the effectiveness of the privacy-preserving multi-teacher distillation framework.

Conclusions:

• The developed framework successfully addresses key challenges in EEG emotion recognition, including data scarcity and privacy.
• This approach shows significant potential for scalable and privacy-preserving emotion recognition applications.