Latent Dynamics · 2026-11-13
Your Vision Foundation Model Is Secretly an EEG Foundation Model
Guoan Wang, Feng Liu
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
SeeEEG converts multichannel EEG into images that a frozen vision transformer can read, by assigning each EEG channel to a dedicated row of ViT tokens and learning only the “display” (montage/filter/contrast) parameters plus a lightweight readout head. Using a frozen DINOv3 ViT-B backbone, the authors report higher balanced accuracy than seven EEG foundation models across eight datasets and strong mean rank improvements in Cohen kappa/AUROC.
Why it matters
The authors argue that EEG foundation models’ frozen features transfer poorly, and that this can be avoided by using a vision foundation model pretrained on natural images. If true, EEG task adaptation can be reduced to learning how EEG is rendered (a small display module) and a head, rather than learning a new encoder from EEG data.
Method
- Keep a pretrained vision backbone frozen (ViT-B) and learn only how EEG is displayed to it (montage, display filter, contrast parameters) plus an identical readout head.
- Render each EEG channel as a dedicated horizontal strip aligned to the ViT patch grid: either a fixed-gain waveform view or a scalogram view with frequency range spanning patch height (task patterns fall within individual tokens).
- Initialize display parameters from a clinical montage; optimize them via task loss through the frozen backbone (at most 1,113 trainable parameters per view). Cross-fit display fitting against evolving readout heads to prevent shortcuts.
Limitation
The paper text does not state a specific limitation.
Abstract (from arXiv)
Electroencephalography (EEG) foundation models are pretrained on thousands of hours of recordings, yet their frozen features transfer poorly to downstream tasks, and this transfer does not consistently improve with larger models or more pretraining data. We show that a vision foundation model that has never seen EEG is a stronger frozen EEG encoder than these specialized models, once the signals are rendered as images that a vision transformer can read. Our method, SeeEEG, keeps the vision backbone frozen and learns only how EEG is displayed to it. Each channel occupies a dedicated row of patches and is drawn either as a waveform at a fixed gain or as a scalogram whose frequency range spans the patch height, so that task-relevant patterns fall within individual tokens. The montage, filtering, and contrast parameters of this display are initialized from a standard clinical montage and optimized with the task loss through the frozen backbone, which amounts to at most 1,113 trainable parameters per view. We compare SeeEEG with seven EEG foundation models on eight clinical, sleep, affective, and brain-computer interface datasets, with every encoder frozen and paired with identical readout heads. With a frozen DINOv3 ViT-B, SeeEEG outperforms every EEG foundation model in balanced accuracy on all eight datasets and reaches a mean rank of 1.3 in Cohen kappa or AUROC, compared with 2.9 for the strongest baseline, REVE. The learned displays transfer without further optimization to the video encoder V-JEPA 2.1, which outperforms every EEG foundation model on both metrics on six of the eight datasets. Replacing DINOv3 with a randomly initialized ViT-B lowers the score by 15.2 points on average, so the ability to read rendered EEG comes from visual pretraining.
Related papers
- Controlled Evaluation of EEG Foundation Models: Random Initialisation, Band-Power and Validation-Selected Probes
- What masking geometry works best for EEG foundation models?
- SPERA: Spherical Prior EEG Foundation Model with Geometry- and Frequency-Aware Latent Prediction
- The (Butter)fly Dream: Linear Separability of World-Model Rollouts in a Connectome-Constrained Visual Model
- Similar Predictive Fit but Different Latent Dynamics: Characterizing Learned Dynamical Structure in Personalized Models of Brain Disorders
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-11-12 · Indexed by Pub-AI: 2026-11-13 UTC
Your Vision Foundation Model Is Secretly an EEG Foundation Model
Preprint posted: 2026-11-12 · Original record ↗
- arxiv
- 2611.17031
Explicit source versions (1)
- v1 · 2026-11-12
- Preprint available
Source date: 2026-11-12 · First detected: 2026-11-13 UTC · Historical baseline
Source evidence (2)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2611.17031v1; arxiv:announce_type=new; item pubDate (official metadata)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Electroencephalography (EEG) foundation models are pretrained on thousands of hours of recordings, yet their frozen features transfer poorly to downstream tasks, and this transfer does not consistently improve with larger models or more pretraining data. We show that a vision foundation model that has never seen EEG is a stronger frozen EEG encoder than these specialized models, once the signals are rendered as images that a vision transformer can read. Our method, SeeEEG, keeps the vision backbone frozen and learns only how EEG is displayed to it. Each channel occupies a dedicated row of patches and is drawn either as a waveform at a fixed gain or as a scalogram whose frequency range spans the patch height, so that task-relevant patterns fall within individual tokens. The montage, filtering, and contrast parameters of this display are initialized from a standard clinical montage and optimized with the task loss through the frozen backbone, which amounts to at most 1,113 trainable parameters per view. We compare SeeEEG with seven EEG foundation models on eight clinical, sleep, affective, and brain-computer interface datasets, with every encoder frozen and paired with identical readout heads. With a frozen DINOv3 ViT-B, SeeEEG outperforms every EEG foundation model in balanced accuracy on all eight datasets and reaches a mean rank of 1.3 in Cohen kappa or AUROC, compared with 2.9 for the strongest baseline, REVE. The learned displays transfer without further optimization to the video encoder V-JEPA 2.1, which outperforms every EEG foundation model on both metrics on six of the eight datasets. Replacing DINOv3 with a randomly initialized ViT-B lowers the score by 15.2 points on average, so the ability to read rendered EEG comes from visual pretraining.
