# What Do EEG Foundation Models Capture from Human Brain Signals?

arXiv: https://arxiv.org/abs/2605.11410 · HTML: https://mlsanity.com/paper/2605.11410/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 17 of its 47 references that are in the map.
- First posted on arXiv 12 May 2026.

## Builds on (most cited first, 17 of 17)

- [Concept Bottleneck Models](https://mlsanity.com/paper/2007.04612/) (2020)
- [Probing Classifiers: Promises, Shortcomings, and Advances](https://mlsanity.com/paper/2102.12452/) (2021)
- [What you can cram into a single vector: Probing sentence embeddings for linguistic properties](https://mlsanity.com/paper/1805.01070/) (2018)
- [Null It Out: Guarding Protected Attributes by Iterative Nullspace Projection](https://mlsanity.com/paper/2004.07667/) (2020)
- [Large Brain Model for Learning Generic Representations with Tremendous EEG Data in BCI](https://mlsanity.com/paper/2405.18765/) (2024)
- [LEACE: Perfect linear concept erasure in closed form](https://mlsanity.com/paper/2306.03819/) (2023)
- [BENDR: using transformers and a contrastive self-supervised learning task to learn from massive amounts of EEG data](https://mlsanity.com/paper/2101.12037/) (2021)
- [CBraMod: A Criss-Cross Brain Foundation Model for EEG Decoding](https://mlsanity.com/paper/2412.07236/) (2024)
- [Linear Adversarial Concept Erasure](https://mlsanity.com/paper/2201.12091/) (2022)
- [NeuroLM: A Universal Multi-task Foundation Model for Bridging the Gap between Language and EEG Signals](https://mlsanity.com/paper/2409.00101/) (2024)
- [Neuro-GPT: Towards A Foundation Model for EEG](https://mlsanity.com/paper/2311.03764/) (2023)
- [CSBrain: A Cross-scale Spatiotemporal Brain Foundation Model for EEG Decoding](https://mlsanity.com/paper/2506.23075/) (2025)
- [EEG2Rep: Enhancing Self-supervised EEG Representation Through Informative Masked Inputs](https://mlsanity.com/paper/2402.17772/) (2024)
- [Kernelized Concept Erasure](https://mlsanity.com/paper/2201.12191/) (2022)
- [Neural network an1alysis of sleep stages enables efficient diagnosis of narcolepsy](https://mlsanity.com/paper/1710.02094/) (2017)
- [Concept-based explainability for an EEG transformer model](https://mlsanity.com/paper/2307.12745/) (2023)
- [An Explainable Deep Learning-Based Method For Schizophrenia Diagnosis Using Generative Data-Augmentation](https://mlsanity.com/paper/2310.16867/) (2023)

## Built on by (most cited first, 3 of 3)

- [The Identity Trap in EEG Foundation Models: A Diagnostic Audit](https://mlsanity.com/paper/2606.06647/) (2026)
- [Cross-Cohort Spectral-Temporal Dissociation in Frozen EEG Foundation-Model Representations](https://mlsanity.com/paper/2607.24834/) (2026)
- [EEG-JEPA: Structured Latent Prediction for EEG Foundation Models](https://mlsanity.com/paper/2608.00114/) (2026)

## Abstract (arXiv)

Clinical electroencephalogram (EEG) analysis rests on a hand-crafted feature catalog refined over decades, e.g., band power, connectivity, complexity, and more. Modern EEG foundation models bypass this catalog, learn directly from raw signals via self-supervised pretraining, and match or outperform feature-engineered baselines on most clinical benchmarks. Whether the two representations align is an open question, which we decompose into three sub-questions: what does the model learn, what does the model use, and how much can be explained. We answer them with layer-wise ridge probing, LEACE-style cross-covariance subspace erasure, and a transparent classifier benchmarked against a random-feature baseline. The audit covers three foundation models (CSBrain, CBraMod, LaBraM), five clinical tasks (MDD, Stress, ISRUC-Sleep, TUSL, Siena), and a 6-family 63-feature lexicon. Of the 945 (model, task, feature) units, 648 (68.6%) are representation-causal and 199 (21.1%) are encoded-only. Across tasks, 50 features qualify as universal candidates with strong support (all three architectures RC) in two or more tasks. Frequency-domain features dominate, but the other five families each contribute substantial causal mass. Confirmed features recover, on average, 79.3% of the foundation model's advantage over the random baseline, with a clean task gradient (MDD ≈ 0.99 down to Stress ≈ 0.56): tasks near ceiling are almost fully recovered by the lexicon, while harder tasks leave a non-trivial residual that pinpoints a concrete target for future concept discovery.