Sequentially Updated Least Squares Support Vector Machine With Applications In Online Brain-Computer Interfaces

Least squares support vector machine (LS-SVM) achieves similar classification performance as conventional SVM by solving a set of linear equations instead of quadratic programming. In this paper, we propose a sequentially updating approach of LS-SVM, which is tailored for online brain-computer interface (BCI) systems where training samples arrive sequentially. Upon each update of the training data set, the sequentially updating approach finds the optimal classifier without matrix inverse operation on the kernal matrix. Hence, it not only reduces the computational load in a significant manner, but also saves the memory for storing past data points. Experimental results show the effectiveness of the proposed approach.