Reducing training requirements through evolutionary based dimension reduction and subject transfer

摘要

Training Brain Computer Interface (BCI) systems to understand the intentionof a subject through Electroencephalogram (EEG) data currently requiresmultiple training sessions with a subject in order to develop the necessaryexpertise to distinguish signals for different tasks. Conventionally the taskof training the subject is done by introducing a training and calibration stageduring which some feedback is presented to the subject. This training sessioncan take several hours which is not appropriate for on-line EEG-based BCIsystems. An alternative approach is to use previous recording sessions of thesame person or some other subjects that performed the same tasks (subjecttransfer) for training the classifiers. The main aim of this study is togenerate a methodology that allows the use of data from other subjects whilereducing the dimensions of the data. The study investigates severalpossibilities for reducing the necessary training and calibration period insubjects and the classifiers and addresses the impact of i) evolutionarysubject transfer and ii) adapting previously trained methods (retraining) usingother subjects data. Our results suggest reduction to 40% of target subjectdata is sufficient for training the classifier. Our results also indicate thesuperiority of the approaches that incorporated evolutionary subject transferand highlights the feasibility of adapting a system trained on other subjects.