Toward metacognition: subject-aware contrastive deep fusion representation learning for EEG analysis

Area of Science:

• Neuroscience
• Machine Learning
• Cognitive Science

Background:

• Accurate classification of confidence levels in visual perception is crucial for understanding cognitive processes.
• Existing methods may not fully leverage the rich information within electroencephalogram (EEG) data or account for inter-subject variability.

Purpose of the Study:

• To develop a subject-aware deep fusion neural network framework, WaveFusion, for precise classification of confidence levels.
• To enhance representation learning and classification accuracy by utilizing subject-specific characteristics in EEG data.

Main Methods:

• WaveFusion employs lightweight convolutional neural networks for time-frequency analysis and an attention network for data integration.
• A subject-aware contrastive learning approach is integrated to leverage multi-subject EEG data heterogeneity.
• The framework performs per-lead time-frequency analysis and integrates modalities for final prediction.

Main Results:

• WaveFusion achieved a high classification accuracy of 95.7% for confidence levels.
• The framework successfully identified influential brain regions related to confidence perception.
• Subject-aware contrastive learning significantly boosted representation learning and classification performance.

Conclusions:

• WaveFusion offers an effective and accurate method for classifying confidence levels in visual perception using EEG data.
• The subject-aware contrastive learning strategy enhances the model's ability to learn robust representations from heterogeneous data.
• The findings highlight the potential of deep learning and EEG analysis in understanding subjective confidence in perception.