Cross-Task Cognitive Workload Recognition Based on EEG and Domain Adaptation.

Area of Science:

• Neuroscience
• Human-Computer Interaction
• Machine Learning

Background:

• Cognitive workload recognition is crucial for operator safety in human-robot interaction.
• Current methods struggle with cross-task workload recognition due to electroencephalogram (EEG) signal discrepancies.
• Generalizing workload models to new tasks remains a significant challenge.

Purpose of the Study:

• To develop and evaluate EEG-based cross-task cognitive workload recognition models.
• To address the challenge of limited generalization in existing workload recognition systems.
• To improve operator health and prevent accidents in human-robot interaction.

Main Methods:

• A fine-grained workload paradigm including working memory and mathematical addition tasks was designed.
• Four domain adaptation methods were explored to bridge task-specific EEG signal discrepancies.
• Support vector machine classifiers were used for low and high workload level classification within a leave-one-task-out cross-validation framework.

Main Results:

• The proposed task transfer framework demonstrated improved performance over non-transfer classifiers, with accuracy gains of 3% to 8%.
• Transfer Joint Matching (TJM) consistently yielded the best results among the evaluated domain adaptation methods.
• The study successfully classified low and high cognitive workload levels across different tasks.

Conclusions:

• Domain adaptation methods are effective in improving cross-task cognitive workload recognition using EEG signals.
• The developed framework enhances the generalizability of workload recognition models, crucial for real-world human-robot interaction.
• This research contributes to safer and more efficient human-robot collaboration by enabling robust workload monitoring.