In machine learning based Brain Computer Interfaces (BCIs), it is a challenge to use only a small amount of labelled data to build a classifier for a specific subject. This challenge was specifically addressed in BCI Competition 2005. Moreover, an effective BCI system should be adaptive to tackle the dynamic variations in brain signal. One of the solutions is to have its parameters adjustable while the system is used online. In this paper we introduce a new semi-supervised support vector machine (SVM) learning algorithm. In this method, the feature extraction and classification are jointly performed in iterations. This method allows us to use a small training set to train the classifier while maintaining high performance. Therefore, the tedious initial calibration process is shortened. This algorithm can be used online to make the BCI system robust to possible signal changes. We analyze two important issues of the proposed algorithm, the robustness of the features to noise and the convergence of algorithm. We applied our method to data from BCI competition 2005, and the results demonstrated the validity of the proposed algorithm
[1]
Rayid Ghani,et al.
Analyzing the effectiveness and applicability of co-training
,
2000,
CIKM '00.
[2]
G. Pfurtscheller,et al.
Brain-Computer Interfaces for Communication and Control.
,
2011,
Communications of the ACM.
[3]
Bernhard Schölkopf,et al.
Learning with Local and Global Consistency
,
2003,
NIPS.
[4]
Sebastian Mika,et al.
Kernel Fisher Discriminants
,
2003
.
[5]
Klaus-Robert Müller,et al.
Boosting bit rates in noninvasive EEG single-trial classifications by feature combination and multiclass paradigms
,
2004,
IEEE Transactions on Biomedical Engineering.
[6]
Gilles Blanchard,et al.
BCI competition 2003-data set IIa: spatial patterns of self-controlled brain rhythm modulations
,
2004,
IEEE Transactions on Biomedical Engineering.