Yumi Dobashi, Atsushi Takemoto, Shu Shigezumi, Takumi Shiraki, Katsuki Nakamura, and Takashi Matsumoto
Biomedical signal processing, biomedical computing, brain–machine interface, electroencephalography
Electroencephalography (EEG) signals are one of the most popular signals used for brain–machine interfaces (BMIs). EEG-based BMI methods often work in batch mode, where a user must conduct the learning phase for a pre-determined period of time. This paper proposes an EEG-based sequential BMI system in which (i) the machine can determine when to end the learning phase automatically by monitoring the learning progress using the sequential error rate (SER) as an evaluation index, and (ii) sequential learning in both the brain and the machine in a cooperative manner is employed. In the proposed approach, called brain–machine co-learning, subjects learn how to use the system by means of real-time visual feedback, whereas the machine learns the subjects’ EEG signals by Bayesian sequential learning. The SER refers to the average classification error rate windowed over a short time period, which was proposed in Hara et al., Sequential error rate evaluation of SSVEP classification Problem with Bayesian sequential learning, Proc. 10th IEEE Int. Conf. on Information Technology and Applications in Biomedicine, Corfu, Greece, November 2–5, 2010, and it represents the status of Bayesian sequential learning in real time. In our proposed approach, subjects can use the system while eliminating unnecessary training. The proposed system was tested against a steady-state visual-evoked potential classification problem. The training phase varied for each subject and was sometimes short, yet satisfactory, leading to high classification accuracy.
Important Links:
Go Back
