Electroencephalography Decoding with Conditional Identification Generator

Area of Science:

• Neuroscience
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Deep neural networks (DNNs) show promise for decoding electroencephalography (EEG) signals in human-AI interaction.
• A key challenge for DNNs is inter-person variability in EEG data, limiting generalizability.

Purpose of the Study:

• To develop a novel framework for enhancing EEG signal decoding by addressing inter-person variability.
• To improve the generalizability and accuracy of DNNs in human-AI interaction systems.

Main Methods:

• Proposed a framework integrating conditional identification information with EEG signals and individual traits.
• Introduced a privacy-preserving generative model to derive embedding knowledge directly from raw EEG signals, avoiding personal identification tests.

Main Results:

• The proposed framework demonstrated superior performance compared to baseline network architectures on the WithMe dataset.
• Achieved substantial improvements in decoding accuracy for both familiar and unseen subjects.

Conclusions:

• The framework offers an efficient, robust, and privacy-conscious approach for human-computer interface systems.
• Leveraging conditional information enhances EEG decoding by accounting for individual differences.