Overview and Benchmarking of EEG Foundation Models (IMAGE)
Caption
Overview of EEG foundation model pre-training and benchmark evaluation. (a) EEG signals are standardized through preprocessing and channel alignment before self-supervised pre-training with different objectives, including raw signal reconstruction, embedded token reconstruction, frequency-domain modeling, codebook learning, and causal modeling. (b) The benchmark covers 13 datasets across nine brain-computer interface paradigms, including motor imagery, P300, SSVEP, clinical EEG detection, emotion recognition, visual decoding, fatigue detection, sleep stage analysis, and workload detection. (c) Two evaluation settings are considered: cross-subject generalization using leave-one-subject-out evaluation and rapid personalization using within-subject few-shot adaptation.
Credit
Dr. Dingkun Liu from Huazhong University of Science and Technology and Zhongguancun Academy, China
Usage Restrictions
Use with credit.
License
Original content