神经科学与认知科学
突破级
暂无讲解视频
收录解读
NeuroAtlas evaluates EEG foundation models across a large collection of clinical EEG and BCI datasets, covering epilepsy, sleep, brain-age estimation, and interface tasks.
The paper argues that generic benchmark metrics are insufficient for clinical utility and introduces task-specific measures such as event-level decisions, hypnogram-derived features, and brain-age gap analyses.
A key finding is skeptical and useful: EEG-specific foundation models do not consistently dominate generic time-series foundation models, and model rankings vary across domains.
For the collection, NeuroAtlas is valuable as a benchmark and evaluation-interface paper that prevents premature claims about unified EEG foundation models.