Integrating metacognitive mechanisms optimizes EEG generative models via hierarchical regularization.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Acquiring sufficient electroencephalography (EEG) data for training deep neural networks (DNNs) in brain-computer interfaces (BCIs) is difficult due to individual variability and data collection limitations like movement artifacts.
• Current methods often use generative models with regularizers to synthesize EEG signals, but selecting effective regularizers is a significant challenge.

Purpose of the Study:

• To develop a novel metacognitive regulation module to improve the generation of realistic and diverse EEG signals for BCI applications.
• To enhance the training of DNNs by addressing the limitations of current EEG data acquisition and generative modeling techniques.

Main Methods:

• Proposed a metacognitive regulation module incorporating three regularizers designed to capture EEG temporal dynamics and functional resolution.
• Validated the module's effectiveness through theoretical analysis and empirical testing on two distinct EEG datasets.

Main Results:

• The metacognitive regulation module significantly enhanced generative models, producing highly complex and realistic EEG activity.
• The module demonstrated improved generalization capabilities across various generative models.
• DNN models trained with the generated data exhibited enhanced human-like decision-making and adaptation abilities.

Conclusions:

• The proposed metacognitive regulation module offers a promising approach to overcome EEG data scarcity in BCI research.
• This method improves the quality and utility of generated EEG data, leading to more robust and capable DNN-based BCIs.