Tzu-Kuo Huang Email: b89034@csie.ntu.edu.tw Kai-Min Chung Email: b88061@csie.ntu.edu.tw Chih-Jen Lin Email: cjlin@csie.ntu.edu.tw Computer Science and Information Engineering National Taiwan University, Taipei, Taiwan Description: Since the training data are signals, they may consist of some noise. Thus, we filter the signals by windowed-fft (low pass at 5Hz). Furthermore, to prevent overfitting the dataset, which has only few training trials comparing to the number of the features, we choose the last 50ms(10 time points) of each channel. We use SVM as the classifer for this problem, since we are developers of SVM software. After the above preprocessing, we do model selection on SVM parameters, and submit the result predicted by the best model.