EEG-FM-Compass: progress, benchmarking and future directions for EEG foundation models
- 1Ministry of Education Key Laboratory of Image Processing and Intelligent Control, School of Artificial Intelligence and Automation, Huazhong University of Science and Technology , Wuhan 430074 ,
- 2Zhongguancun Academy , Beijing 100094 ,
- 3State Key Laboratory of Brain Cognition and Brain-inspired Intelligence Technology, Institute of Automation, Chinese Academy of Sciences , Beijing 100190 ,
Abstract
Electroencephalography (EEG) foundation models (FMs) have recently emerged as a promising paradigm for brain–computer interfaces, aiming to learn transferable neural representations from large-scale heterogeneous recordings. Despite rapid progress, a fair and comprehensive comparison of existing EEG FMs is still lacking, owing to inconsistent pre-training objectives, preprocessing choices and downstream evaluation protocols. To fill this gap, we present EEG-FM-Compass. We first review 55 representative models and organize their design choices into a unified taxonomic framework including data standardization, model architectures and self-supervised pre-training strategies. We then evaluate 12 open-source FMs and competitive specialist baselines across 13 EEG datasets spanning 9 brain–computer interface paradigms. Emphasizing real-world deployments, we consider both cross-subject generalization under a leave-one-subject-out protocol and rapid calibration under a within-subject few-shot setting. We further compare full-parameter fine-tuning with linear probing to assess the transferability of pre-trained representations, and examine the relationship between model scale and downstream performance. Our results indicate that (i) linear probing is frequently insufficient; (ii) specialist models trained from scratch remain competitive across many tasks; and (iii) larger FMs do not necessarily yield better generalization performance under current data regimes and training practices.
Related
- Dendritic cell-based vaccines for colorectal cancer: Mechanisms, clinical progress and future directions (Review)
- CRISPR‐based therapeutic and modelling approaches in Huntington's disease: Progress, challenges and future directions
- Decision benchmarking reveals when protein foundation models add experimental value
- Vagus nerve stimulation for trigeminal autonomic cephalalgias: Evidence, challenges and future directions
- Cattle production, water resources, and sustainability: a review of current knowledge and future directions.
BibTeX
@article{liu2026eegfmcompass,
title = {{EEG-FM-Compass: progress, benchmarking and future directions for EEG foundation models}},
author = {Dingkun Liu and Yuheng Chen and Zhu Chen and Zhenyao Cui and Yaozhi Wen and Jiayu An and Jingwei Luo and Dongrui Wu},
journal = {National Science Review},
year = {2026},
month = jul,
doi = {10.1093/nsr/nwag466},
url = {https://doi.org/10.1093/nsr/nwag466},
note = {CC BY},
}
RIS
TY - JOUR
TI - EEG-FM-Compass: progress, benchmarking and future directions for EEG foundation models
AU - Dingkun Liu
AU - Yuheng Chen
AU - Zhu Chen
AU - Zhenyao Cui
AU - Yaozhi Wen
AU - Jiayu An
AU - Jingwei Luo
AU - Dongrui Wu
PY - 2026
DA - 2026/07/31
JO - National Science Review
DO - 10.1093/nsr/nwag466
UR - https://doi.org/10.1093/nsr/nwag466
AB - Electroencephalography (EEG) foundation models (FMs) have recently emerged as a promising paradigm for brain–computer interfaces, aiming to learn transferable neural representations from large-scale heterogeneous recordings. Despite rapid progress, a fair and comprehensive comparison of existing EEG FMs is still lacking, owing to inconsistent pre-training objectives, preprocessing choices and downstream evaluation protocols. To fill this gap, we present EEG-FM-Compass. We first review 55 representative models and organize their design choices into a unified taxonomic framework including data standardization, model architectures and self-supervised pre-training strategies. We then evaluate 12 open-source FMs and competitive specialist baselines across 13 EEG datasets spanning 9 brain–computer interface paradigms. Emphasizing real-world deployments, we consider both cross-subject generalization under a leave-one-subject-out protocol and rapid calibration under a within-subject few-shot setting. We further compare full-parameter fine-tuning with linear probing to assess the transferability of pre-trained representations, and examine the relationship between model scale and downstream performance. Our results indicate that (i) linear probing is frequently insufficient; (ii) specialist models trained from scratch remain competitive across many tasks; and (iii) larger FMs do not necessarily yield better generalization performance under current data regimes and training practices.
SN - 2053-714X
SN - 2095-5138
ER -