Kim Keun Tae Choi Junhyuk KIM HYUNG MIN Lee, Song Joo 2024-01-12T04:08:21Z 2024-01-12T04:08:21Z 2021-12-14 2021-02-22 https://pubs.kist.re.kr/handle/201004/77767 ―Nowadays, pattern recognition or machine learning-based various approaches have been applied to the steady-state somatosensory evoked potential (SSSEP)-based brain-computer interfaces (BCIs) for improving the accuracy of classifying the user’s selective attention to vibration stimulation. However, the accuracy still needs to be improved for applying to a real-world environment. In this paper, the subject-transfer strategy based on the convolutional neural network (CNN) was investigated, if it can improve classification accuracy within SSSEP-based BCIs. In our experiment, for the subjecttransferring, the supportive subject’s data and target subject's data were used to training the common spatial pattern (CSP) for feature extraction and CNN for classification. The SSSEPs were acquired by selective attention to one of the vibration motors (left-hand and right-hand). The experimental results based on 10-fold cross-validation indicate that the subject-transfer method showed approximately 5% higher accuracies than the traditional self-training method, which is training the CSP and CNN using only the target subject’s data. Based on these results, we can confirm that the subject-transfer strategy can contribute to improving the accuracy of the SSSEP-based BCIs for realworld applications. English IEEE Subject-transfer Approach Convolutional Neural Network Steady-state somatosensory evoked potential Brain-Computer Interfaces Subject-Transfer Approach based on Convolutional Neural Networks for the SSSEP-BCIs Conference 1 2021 9th International Winter Conference on Brain-Computer Interface 2021 9th International Winter Conference on Brain-Computer Interface KO high1 resort, Korea 2021-02-22 International Winter Conference on Brain-Computer Interface