# Mechanistic Interpretability of EEG Foundation Models via Sparse Autoencoders

arXiv: https://arxiv.org/abs/2605.13930 · HTML: https://mlsanity.com/paper/2605.13930/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in Interpreting biomedical and science models, one of 13 directions in Mechanistic interpretability.
- Builds on 20 of its 29 references that are in the map.
- First posted on arXiv 13 May 2026.

## Placed in

- Mechanistic interpretability / [Interpreting biomedical and science models](https://mlsanity.com/mechinterp/interpreting-biomedical-and-science-models/)

## Builds on (most cited first, 20 of 20)

- [Representation Learning with Contrastive Predictive Coding](https://mlsanity.com/paper/1807.03748/) (2018)
- [Masked Autoencoders Are Scalable Vision Learners](https://mlsanity.com/paper/2111.06377/) (2021)
- [BEiT: BERT Pre-Training of Image Transformers](https://mlsanity.com/paper/2106.08254/) (2021)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Understanding intermediate layers using linear classifier probes](https://mlsanity.com/paper/1610.01644/) (2016)
- [Toy Models of Superposition](https://mlsanity.com/paper/2209.10652/) (2022)
- [Probing Classifiers: Promises, Shortcomings, and Advances](https://mlsanity.com/paper/2102.12452/) (2021)
- [Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2](https://mlsanity.com/paper/2408.05147/) (2024)
- [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)
- [k-Sparse Autoencoders](https://mlsanity.com/paper/1312.5663/) (2013)
- [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)
- [REVE: A Foundation Model for EEG -- Adapting to Any Setup with Large-Scale Pretraining on 25,000 Subjects](https://mlsanity.com/paper/2510.21585/) (2025)
- [InterPLM: Discovering Interpretable Features in Protein Language Models via Sparse Autoencoders](https://mlsanity.com/paper/2412.12101/) (2024)
- [Concept-based explainability for an EEG transformer model](https://mlsanity.com/paper/2307.12745/) (2023)
- [SPEED: Scalable Preprocessing of EEG Data for Self-Supervised Learning](https://mlsanity.com/paper/2408.08065/) (2024)
- [Mammo-SAE: Interpreting Breast Cancer Concept Learning with Sparse Autoencoders](https://mlsanity.com/paper/2507.15227/) (2025)
- [MedSAE: Dissecting MedCLIP Representations with Sparse Autoencoders](https://mlsanity.com/paper/2510.26411/) (2025)
- [Pretraining on Sleep Data Improves non-Sleep Biosignal Tasks](https://mlsanity.com/paper/2605.02500/) (2026)
- [Clinical Feasibility of Smartphone-based EEG in Kenya](https://mlsanity.com/paper/2605.08157/) (2026)

## Built on by (most cited first, 5 of 5)

- [Cross-Cohort Spectral-Temporal Dissociation in Frozen EEG Foundation-Model Representations](https://mlsanity.com/paper/2607.24834/) (2026)
- [ECG-InterpBench: Benchmarking the Interpretability of ECG Foundation Models with Matched-Scale Sparse Autoencoders](https://mlsanity.com/paper/2607.27404/) (2026)
- [Interpretable MEG Decoding of Perceived Speech: Cortical Sources and the Stimulus Features That Drive Retrieval](https://mlsanity.com/paper/2608.01481/) (2026)
- [Learning transferable human physiology from two million hours of sleep with SleepFM-2](https://mlsanity.com/paper/2609.06849/) (2026)
- [Clinical Concept Centers in LLMs](https://mlsanity.com/paper/2610.02829/) (2026)

## Abstract (arXiv)

EEG foundation models achieve state-of-the-art clinical performance, yet the internal computations driving their predictions remain opaque: a barrier to clinical trust. We apply TopK Sparse Autoencoders (SAEs) across three architecturally distinct EEG transformers: SleepFM, REVE, and LaBraM to extract sparse feature dictionaries from their embeddings. By grounding these features in a clinical taxonomy (abnormality, age, sex, and medication), we benchmark monosemanticity and entanglement across architectures. A single hyperparameter procedure, driven by an intrinsic dictionary health audit, transfers robustly across all three architectures. Via concept steering, we introduce a "target vs. off-target" probe area metric to quantify steering selectivity and reveal three operational regimes: selectively steerable, encoded but entangled, and non-encoded. This framework exposes critical representational failures: "wrecking-ball" interventions that collapse global model performance, and clinical entanglements, such as age-pathology confounding, where it is impossible to suppress one concept without corrupting the other. Finally, a spectral decoder maps these interventions back to the amplitude spectrum, translating latent manipulations into physiologically interpretable frequency signatures, such as pathological slow-wave suppression and α-band restoration.