Optimization of transfer learning based on source sample selection in Euclidean space for P300-based brain-computer...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalogram (EEG) signals, particularly event-related potentials (ERPs) like P300, are crucial for brain-computer interfaces (BCIs).
• Real-time P300-based BCIs face challenges due to EEG signal non-stationarity and inter-subject data variability, necessitating extensive calibration and training data.

Purpose of the Study:

• To develop an efficient transfer learning approach for P300 detection in BCIs.
• To overcome limitations of non-stationary EEG signals and reduce the need for large training datasets.

Main Methods:

• A convolutional neural network (CNN) was employed for high-level feature extraction.
• Euclidean space data alignment and a source selection technique were used to harmonize feature distributions between source and target domains.
• A discriminative restricted Boltzmann machine (RBM) served as the classifier for P300 detection.

Main Results:

• The proposed method achieved an average accuracy of 97% on benchmark datasets (BCI Competition III dataset II and RSVP).
• The technique requires less than half the training samples compared to previous studies.
• Performance was comparable to state-of-the-art methods in both online and offline evaluations.

Conclusions:

• The developed transfer learning technique provides an efficient solution for robust ERP-based BCIs.
• This method significantly reduces the amount of training data required, making BCIs more practical.
• The approach demonstrates strong performance despite reduced training data, addressing a key challenge in BCI development.