A Transfer Learning Algorithm to Reduce Brain-Computer Interface Calibration Time for Long-Term Users

Current motor imagery-based brain-computer interface (BCI) systems require a long calibration time at the beginning of each session before they can be used with adequate levels classification accuracy. In particular, this issue significant burden for term BCI users. This article proposes novel transfer learning algorithm, called r-KLwDSA, to reduce long-term The proposed r-KLwDSA algorithm aligns user's EEG data collected in previous sessions few trials current session, using linear alignment method. Thereafter, aligned from and are fused through weighting mechanism calibrating model. To validate large dataset containing 11 stroke patients, performing 18 sessions, was used. framework demonstrated improvement accuracy, over 4% compared session-specific when there were as two per class available session. particularly successful improving accuracy that had initial below 60%, an average around 10% leading more patients having meaningful rehabilitation.