TY - JOUR
T1 - Toward consistency between humans and classifiers
T2 - Improved performance of a real-time brain–computer interface using a mutual learning system
AU - Lin, Chun Yi
AU - Lu, Chia Feng
AU - Jao, Chi Wen
AU - Wang, Po Shan
AU - Wu, Yu Te
N1 - Publisher Copyright:
© 2023 Elsevier Ltd
PY - 2023/9/15
Y1 - 2023/9/15
N2 - The performance of electroencephalography (EEG) classifiers in a brain–computer interface (BCI) depends heavily on the quality and consistency of training data. Therefore, facilitating collaboration between two independent systems, namely humans and machines, as well as increasing model generalizability and personalization are important tasks. In this study, we designed a mutual learning system to stabilize the EEG patterns of users who performed motor imagery (MI) and attention tasks, and we updated the parameters of a deep learning classifier in real time to improve consistency between the system and users. According to our results, the accuracy of the users on the MI task increased from 56% ± 13.9% to 81.5% ± 8.18%, and that on the attention task increased from 55% ± 7.07% to 82.5% ± 12.3% after application of the proposed mutual learning system. Thus, the mutual learning system facilitates the application of personalized BCIs and heralds a new era in which humans and machines can learn from each other.
AB - The performance of electroencephalography (EEG) classifiers in a brain–computer interface (BCI) depends heavily on the quality and consistency of training data. Therefore, facilitating collaboration between two independent systems, namely humans and machines, as well as increasing model generalizability and personalization are important tasks. In this study, we designed a mutual learning system to stabilize the EEG patterns of users who performed motor imagery (MI) and attention tasks, and we updated the parameters of a deep learning classifier in real time to improve consistency between the system and users. According to our results, the accuracy of the users on the MI task increased from 56% ± 13.9% to 81.5% ± 8.18%, and that on the attention task increased from 55% ± 7.07% to 82.5% ± 12.3% after application of the proposed mutual learning system. Thus, the mutual learning system facilitates the application of personalized BCIs and heralds a new era in which humans and machines can learn from each other.
KW - Bio-feedback
KW - Brain–computer interface
KW - Convolutional neural network
KW - End-to-end learning
KW - Mutual learning system
UR - http://www.scopus.com/inward/record.url?scp=85153682566&partnerID=8YFLogxK
U2 - 10.1016/j.eswa.2023.120205
DO - 10.1016/j.eswa.2023.120205
M3 - Article
AN - SCOPUS:85153682566
SN - 0957-4174
VL - 226
JO - Expert Systems with Applications
JF - Expert Systems with Applications
M1 - 120205
ER -