Online Privacy-Preserving EEG Classification by Source-Free Transfer Learning.

Area of Science:

• Neuroscience and Machine Learning
• Brain-Computer Interface (BCI) Technology

Background:

• Electroencephalogram (EEG) signals are crucial for BCI applications, with transfer learning improving new user adaptation.
• EEG data privacy is a significant concern, and current methods often require new subject data for adaptation.
• Practical BCI deployment is hindered by the need for extensive data collection and calibration for new users.

Purpose of the Study:

• To propose a privacy-preserving method for EEG classification using transfer learning.
• To address the challenge of limited or no new subject data availability in practical BCI scenarios.
• To enable prompt BCI utilization while safeguarding sensitive EEG data.

Main Methods:

• Introduced Online Source-Free Transfer Learning (OSFTL) with distinct offline and online stages.
• Offline stage: Extracts source model parameters from multiple subjects, preserving source data privacy.
• Online stage: Trains a target classifier on sequential EEG data and combines source/target classifiers, dynamically adjusting weights for optimal transferability.

Main Results:

• OSFTL effectively classifies EEG signals while preserving privacy.
• The method demonstrated strong performance in both simulated and real-world BCI applications.
• Dynamic weight adjustment improved the transferability of knowledge from source to target domains.

Conclusions:

• OSFTL offers an effective solution for privacy-preserving EEG classification in BCI.
• The approach facilitates rapid BCI deployment by eliminating the need for new subject data collection.
• OSFTL shows significant potential for advancing BCI applications beyond controlled laboratory environments.