Noise-Factorized Disentangled Representation Learning for Generalizable Motor Imagery EEG Classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor Imagery (MI) Electroencephalography (EEG) is a key Brain-Computer Interface (BCI) paradigm.
• Current MI EEG algorithms struggle with poor generalization across subjects and sessions, limiting real-world BCI applications.

Purpose of the Study:

• To develop a novel framework to improve the generalization capability of MI EEG classification.
• To disentangle EEG data into subject/session-specific, task-specific, and noise components.

Main Methods:

• A joint discriminative and generative framework was proposed.
• The framework utilizes fundamental training losses and strategies to disentangle EEG representations.
• The approach was evaluated on three public MI EEG datasets.

Main Results:

• The proposed framework demonstrated superior performance compared to state-of-the-art benchmark algorithms.
• The method significantly improved generalization across different subjects and sessions.
• Detailed experimental results confirmed the effectiveness of the approach.

Conclusions:

• The novel framework effectively enhances the generalization of MI EEG classification systems.
• Disentangling EEG data into specific components is a promising strategy for robust BCI development.
• This approach paves the way for more reliable and widely applicable BCI technologies.