EEG-Arena

EEG-Arena is an open-source benchmark for systematically evaluating 30 EEG foundation models and 25 supervised baselines across 57 downstream tasks, 23 public datasets, and five experimental protocols.

Download DataarXivCodeLeaderboard

A standardized, large-scale benchmark software package.

EEG-Arena is an open-source benchmark covering 30 EEG FMs and 25 strong supervised baselines across 57 downstream tasks from 23 datasets spanning seven EEG paradigms.

Download our Annotated downstream datasets to quickly run evaluations across multiple settings. You can also quickly adapt your own model to EEG-Arena and evaluate its performance.

EEG-Arena benchmark software package, benchmarked model families, and downstream datasets and tasks across seven EEG paradigms.
Figure 1a-b. The standardized EEG-Arena software package and its large-scale coverage of benchmark models, datasets, tasks, and EEG paradigms.

Large-scale evaluation from five protocols.

EEG-Arena evaluates and compares EEG FMs with strong supervised baselines from multiple perspectives, then brings the evidence together through a multidimensional evaluation framework.

Explore detailed results for Protocol I, Protocol II, Protocol III, and Protocol IV.

Five EEG-Arena evaluation protocols covering comparative performance, pretraining effectiveness, model scaling, pretraining data scaling, and channel robustness.
Figure 1c. Five complementary protocols examine comparative performance, pretraining effectiveness, model scaling, pretraining data scaling, and robustness to channel configuration.

Upload your model results and expand the benchmarks.

EEG-Arena provides developer-friendly interfaces for quickly integrating new models and evaluating them across a broad range of downstream tasks.

We are actively developing NeuroInfra to support uploading evaluation results, expanding benchmark datasets and tasks, and continuously updating model results. Follow our progress and share feedback.