Novel Transfer Learning Approaches forImproving Brain Computer Interfaces

Despite several recent advances, most of the electroencephalogram(EEG)-based brain-computer interface (BCI) applications are still limited to the laboratory due to their long calibration time. Due toconsiderable inter-subject/inter-session and intra-session variations, atime-consuming and fatiguing calibration phase is typically conductedat the beginning of each new session to acquire sufficient labelled train-ing data to train the subject-specific BCI model.This thesis focuses on developing reliable machine learning algorithmsand approaches that reduce BCI calibration time while keeping accu-racy in an acceptable range. Calibration time could be reduced viatransfer learning approaches where data from other sessions or sub-jects are mined and used to compensate for the lack of labelled datafrom the current user or session. In BCI, transfer learning can beapplied on either raw EEG, feature or classification domains.In this thesis, firstly, a novel weighted transfer learning approach isproposed in the classification domain to improve the MI-based BCIperformance when only few subject-specific trials are available fortraining.Transfer learning techniques should be applied in a different domainbefore the classification domain to improve the classification accuracyfor subjects whom their subject-specific features for different classesare not separable. Thus, secondly, this thesis proposes a novel regu-larized common spatial patterns framework based on dynamic timewarping and transfer learning (DTW-R-CSP) in raw EEG and featuredomains.In previous transfer learning approaches, it is hypothesised that thereare enough labelled trials available from the previous subjects or ses-sions. However, in the case when there are no labelled trials available from other subjects or sessions, domain adaptation transfer learningcould potentially mitigate problems of having small training size byreducing variations between the testing and training trials. Thus, todeal with non-stationarity between training and testing trials, a novelensemble adaptation framework with temporal alignment is proposed.