EEG Foundation Model Representations Under Hierarchical Classification and Short Calibration

Key finding. Frozen and calibrated on ~58 epochs of the user, foundation-model representations do not beat classical baselines. Adapted on other subjects, they barely change detection but lift lateralization: fine-tuned REVE reaches κ = 0.445 with 8 electrodes, against 0.283 for CSP.

Links

• Poster (PDF)90 × 120 cm, 2.3 MB
• Contact Eduardaeduarda.tpereira@inf.ufrgs.br
• MLAB · INF-UFRGSMachine Learning & Applications in Biomedical Data Lab
• DatasetPhysioNet EEG Motor Movement/Imagery
• Previous workScoping review of ML for hand motor imagery, ERAMIA-RS 2025
• CBraModcode and pretrained weights
• LUNAcode and pretrained weights (BioFoundation)
• REVEproject page, code and pretrained weights

References

Numbered as on the poster.

1. 1
   Bitencourt, F. C. et al. (2025). The Landscape of Machine Learning for EEG Hand Motor Imagery Decoding: A Scoping Review (2020–2025). Anais da I ERAMIA-RS, SBC, 408–411.

   SBC OpenLib
2. 2
   El Ouahidi, Y. et al. (2025). REVE: A Foundation Model for EEG — Adapting to Any Setup with Large-Scale Pretraining on 25,000 Subjects. arXiv:2510.21585.

   arXivproject page
3. 3
   Wang, J. et al. (2025). CBraMod: A Criss-Cross Brain Foundation Model for EEG Decoding. ICLR.

   arXivcode
4. 4
   Döner, B. et al. (2025). LUNA: Efficient and Topology-Agnostic Foundation Model for EEG Signal Analysis. NeurIPS.

   arXivcode
5. 5
   Yao, Y. et al. (2025). Foundation models for EEG decoding: current progress and prospective research. J. Neural Eng., 22(6).

   DOI
6. 6
   Sarhane, A. et al. (2026). Stacked LoRA for Subject-Adaptive EEG Foundation Models in Motor Imagery Decoding. arXiv:2607.03094.

   arXiv
7. 7
   Schalk, G. et al. (2004). BCI2000: A General-Purpose Brain-Computer Interface (BCI) System. IEEE Trans. Biomed. Eng., 51(6), 1034–1043. Schalk, G. (2009). EEG Motor Movement/Imagery Dataset (v1.0.0). PhysioNet. Pollard, T. et al. (2026). PhysioNet as a Global Platform for Biomedical Research. Nature Health.

   BCI2000 DOIdatasetNature Health DOI
8. 8
   Ramoser, H. et al. (2000). Optimal spatial filtering of single trial EEG during imagined hand movement. IEEE Trans. Rehabil. Eng., 8(4), 441–446.

   DOI
9. 9
   Pfurtscheller, G., Lopes da Silva, F. H. (1999). Event-related EEG/MEG synchronization and desynchronization: basic principles. Clin. Neurophysiol., 110(11), 1842–1857.

   DOI
10. 10
    Yang, L. et al. (2026). Are EEG Foundation Models Worth It? Comparative Evaluation with Traditional Decoders in Diverse BCI Tasks. ICLR.

    ICLR page