65. towards plug-and-play brain-state decoding with large-scale data
Name: Chunshu Wei
Grad Year: 2017
Pervasive and elusive variability of human brain dynamics poses significant challenges for developing practical real-world brain-state decoding with non-invasive electroencephalogram (EEG). Conventional individualized decoding model requires time-consuming and expensive training session to acquire sufficient task-related data before each use. In this study, we proposed a subject-transfer framework that leverages large-scale existing data to overcome the inter- and intra-subject variability in state-related EEG responses. Hierarchical clustering analysis was applied to quantitatively visualize and analyze the variability across sessions and subjects within the dataset, and established the fundamental of the subject-transfer framework. Next, we used easily-collected task-free brain activity to estimate subject similarity for the efficiency of transferring models among subjects. Finally, we investigated the relationship between the size of data and the subject-transferring performance comparing to traditional self-decoding approaches. The proposed subject-transfer framework successfully reduce the calibration time for EEG-based drowsiness detection from ~90 minutes to ~100 seconds while maintaining comparable accuracy. This demonstration would considerably facilitate further exploration of plug-and-play brain decoding for real-world applications of brain-computer interfaces.
Industry Application Area(s)
Internet, Networking, Systems | Life Sciences/Medical Devices & Instruments | Software, Analytics