49. TRANSFER LEARNING WITH LARGE-SCALE DATA IN BRAIN-COMPUTER INTERFACES
Name: Chunshu Wei
Grad Year: 2017
Brain-computer Interfaces (BCIs) have been developed for translating specific patterns of brain activities into comprehensible commands to control computers or external devices. To deal with individual differences in human electroencephalogram (EEG), BCIs often require a significant amount of training data to build and calibrate a reliable model for each individual. This user-specific training/calibration is not only labor intensive and time consuming, but also hinders the applications of BCIs in real life. To alleviate this problem, transfer learning (TL) has been employed to leverage existing data from other sessions or subjects to build a BCI for a new user with limited calibration data. However, the TL approaches still require representative training data under each of conditions to be classified, which might be problematic when the data of one or more conditions are difficult or expensive to obtain. This study proposed a novel TL framework that could leverage large-scale existing data from other subjects and a very limited amount of calibration data from the test subject. This study also demonstrated the efficacy of this method through a BCI that detected lapses during driving. With the help of large-scale existing data, the proposed TL approach outperformed the within-subject approach while considerably reducing the required calibration data for the target subject (only ~1.5 min of data from each individual as opposed to ~90 min of a pilot session used in the within-subject approach). The TL approach can enable and facilitate numerous real-world applications (not limited to lapse detection) of BCIs.
Industry Application Area(s)
Life Sciences/Medical Devices & Instruments | Software, Analytics