Multi-Task Collaborative Network: Bridge the Supervised and Self-Supervised Learning for EEG Classification in RSVP...

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Electroencephalography (EEG) data in rapid serial visual presentation (RSVP) tasks suffer from low signal-to-noise ratios and noisy labels, limiting classification accuracy.
• Traditional supervised learning (SL) approaches often lead to overfitting and poor generalization in these challenging datasets.

Purpose of the Study:

• To develop a novel multi-task collaborative network (MTCN) for enhanced EEG representation learning in RSVP tasks.
• To integrate both supervised learning (SL) and self-supervised learning (SSL) to improve classification performance and generalization.

Main Methods:

• The proposed MTCN utilizes the primary RSVP EEG classification task (SL) for initial representation and thresholding.
• Two self-supervised learning (SSL) tasks, masked temporal and spatial recognition, are incorporated to refine temporal dynamics and spatial relationships.
• Features are explicitly decomposed into task-specific and task-shared components to foster collaboration between SL and SSL.

Main Results:

• The MTCN effectively extracts more generalized EEG representations by simultaneously learning from multiple tasks.
• Experiments on THU, CAS, and GIST datasets demonstrate significant improvements in RSVP task performance.
• The approach mitigates overfitting and enhances generalization capabilities compared to traditional methods.

Conclusions:

• The multi-task collaborative network (MTCN) offers a robust framework for improving EEG analysis in RSVP tasks.
• Integrating SL and SSL through MTCN leads to superior feature extraction and classification performance.
• The method provides a promising direction for handling noisy and low-signal EEG data.