Weighted multi-task learning in classification domain for improving brain-computer interface

One of the major limitations of brain computer interface (BCI) is its long calibration time. Due to between sessions/subjects nonstationarity, typically a big amount of training data needs to be collected at the beginning of each session in order to tune the parameters of the system for the target user. In this paper, a number of novel weighted multi-task transfer learning algorithms are proposed in the classification domain to reduce the calibration time without sacrificing the classification accuracy of the BCI system. The proposed algorithms use data from other subjects and combine them to estimate the classifier parameters for the target subject. This combination is done based on how similar the data from each subject is to the few trials available from the target subject. The proposed algorithms are evaluated using dataset 2a from BCI competition IV. According to the results, the proposed algorithms lead to reduce the calibration time by 75% and enhance the average classification accuracy at the same time.