Regularized RKHS-Based Subspace Learning for Motor Imagery Classification.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interface (BCI) technology offers communication pathways for individuals with disabilities.
• Non-invasive electroencephalogram (EEG) signals are promising but challenging due to their non-stationary nature.
• Domain adaptive learning (DAL) shows potential for handling time-varying signals.

Purpose of the Study:

• To develop a novel regularized reproducing kernel Hilbert space (RKHS) subspace learning algorithm for motion imagery classification.
• To address the limitations of existing methods, particularly the Maximum Mean Difference (MMD) criterion, by incorporating local distribution information.
• To improve the adaptability and effectiveness of BCIs for real-world applications.

Main Methods:

• A regularized RKHS subspace learning algorithm is proposed, incorporating K-nearest neighbors (KNNs) for classification.
• A novel regularization term from source domain linear discriminant analysis (SLDA) is introduced to optimize source domain data distribution.
• The RKHS subspace framework is constructed sparsely to account for BCI data sensitivity.

Main Results:

• The proposed SLDA regularization improved baseline algorithms' average accuracy by 2-9% on standard datasets.
• The algorithm achieved 3% higher average accuracy in motion imagery classification compared to other methods.
• Experimental results demonstrate the algorithm's adaptability and effectiveness in handling non-stationary EEG signals.

Conclusions:

• The proposed regularized RKHS subspace learning algorithm effectively addresses the challenges of non-stationary EEG signals in BCIs.
• The integration of SLDA offers a significant improvement over traditional MMD criteria for distribution adaptation.
• This work advances BCI technology, enhancing classification accuracy and user experience for individuals with motor impairments.