Research work · 2010
Multitask Learning for Brain-Computer Interfaces
Morteza Alamgir, Moritz Grosse‐Wentrup, Yasemin Altün
Abstract
Brain-computer interfaces (BCIs) are lim-ited in their applicability in everyday settings by the current necessity to record subject-specific calibration data prior to actual use of the BCI for communication. In this pa-per, we utilize the framework of multitask learning to construct a BCI that can be used without any subject-specific calibration pro-cess. We discuss how this out-of-the-box BCI can be further improved in a computation-ally efficient manner as subject-specific data becomes available. The feasibility of the ap-proach is demonstrated on two sets of exper-imental EEG data recorded during a stan-dard two-class motor imagery paradigm from a total of 19 healthy subjects. Specifically, we show that satisfactory classification re-sults can be achieved with zero training data, and combining prior recordings with subject-specific calibration data substantially outper-forms using subject-specific data only. Our results further show that transfer between recordings under slightly different experimen-tal setups is feasible. 1