Data set IIb: P300 Speller Group information: Department of Biomedical Engineering Tsinghua University Beijing, 100084, P. R. China Supervisor: GAO Xiaorong, Ph.D. ( gxr-dea@tsinghua.edu.cn ) Team Member: XU Neng, MIAO Xiaobo, HONG Bo Counselor: GAO Shangkai, YANG Fusheng Advanced results With the 15 sequences of each word, our results are: FOOD MOOT HAM PIE CAKE TUNA ZYGOT 4567 Using our algorithm, we can get the same results with only the first 8 sequences for each word, and with the sequences from 5 to 9, that is to say, only using these 5 consecutive sequences, we still can get the same results. We use 3 different methods to predict the words. Here, "Get the same results" means all 3 methods get the same results when using the same consecutive sequences with the specific numbers of sequence. In a word, our algorithm is robust for only using 8 sequences, and sometimes for 5 consecutive sequences. Algorithm. Our algorithm is indicated in the flowchart below, with the training part in the frame and testing part out of the frame. For this kind of offline prediction, training part and testing part do not work simultaneously. Training parts require one or several minutes (Pentium III 866MHz, 256M SDRAM, Windows XP, MATLIB 6.1), but for each person, training parts are needed to be executed only once, after that the parameters are stored and are used for all the testing parts. Thus, after training, testing part can be implemented in real time. [Figure: see DOC file]