EXPOSE: Explainable and Domain-Robust Embeddings from Pathology Vision Foundation Models using Sparse Autoencoders
Not on the map. Why?
Abstract (arXiv)
Vision Foundation Models (VFMs) are widely used in computational pathology but remain sensitive to domain shifts arising from variations in staining, tissue preparation, and scanner hardware. A key limitation is that VFM embeddings entangle biological with domain-specific information, hindering cross-domain generalization. We propose Explainable Probing of Cross-Domain Sparse Embeddings (EXPOSE), a framework that uses Sparse Autoencoders (SAEs) as an explainable bottleneck to identify and suppress domain-specific components in VFM embeddings. We train a sparse representation of VFM features, use a linear classifier to identify domain-specific latent dimensions, and mask these features prior to downstream relapse prediction without retraining the backbone model. Experiments on a large prostate cancer dataset with multiple acquisition domains show that SAE features capture both domain- and task-specific information, which are partially disentangled in the latent space. Removing domain-specific features improves cross-domain performance and increases embedding robustness as measured by the Domain Robustness Index (DoRI). Code is available at https://github.com/imsb-uke/expose .
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 25 of its 41 references that are in the map. First posted on arXiv 28 Aug 2026.
Builds on
25 of 41 references.
References · 25
- 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,359
- 2024Scaling and evaluating sparse autoencoderscited by 530
- 2021Domain Adaptation for Medical Image Analysis: A Surveycited by 351
- 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoderscited by 283
- 2024Improving Dictionary Learning with Gated Sparse Autoencoderscited by 171
- 2024Virchow2: Scaling Self-Supervised Mixed Magnification Models in Pathologycited by 144
- 2025Sparse Autoencoders Learn Monosemantic Features in Vision-Language Modelscited by 70
- 2024A Clinical Benchmark of Public Self-Supervised Pathology Foundation Modelscited by 64
- 2024Hibou: A Family of Foundational Vision Transformers for Pathologycited by 52
- 2025Current Pathology Foundation Models are unrobust to Medical Center Differencescited by 37
- 2025Distilling foundation models for robust and efficient models in digital pathologycited by 33
- 2025Steering CLIP's vision transformer with sparse autoencoderscited by 33
13 more
- 2025SAE-SSV: Supervised Steering in Sparse Representation Spaces for Reliable Control of Language Modelscited by 28
- 2025Do Multiple Instance Learning Models Transfer?cited by 28
- 2025Interpretable and Testable Vision Features via Sparse Autoencoderscited by 24
- 2025Foundation Models in Computational Pathology: A Review of Challenges, Opportunities, and Impactcited by 22
- 2025A Survey on Computational Pathology Foundation Models: Datasets, Adaptation Strategies, and Evaluation Taskscited by 15
- 2025A Survey of Pathology Foundation Model: Progress and Future Directionscited by 14
- 2024Beyond Interpretability: The Gains of Feature Monosemanticity on Model Robustnesscited by 13
- 2025CytoSAE: Interpretable Cell Embeddings for Hematologycited by 8
- 2025Pathology Foundation Models are Scanner Sensitive: Benchmark and Mitigation with Contrastive ScanGen Losscited by 7
- 2025Analyzing Hierarchical Structure in Vision Models with Sparse Autoencoderscited by 5
- 2025Prototype-Based Multiple Instance Learning for Gigapixel Whole Slide Image Classificationcited by 2
- 2021Towards Explainable End-to-End Prostate Cancer Relapse Prediction from H&E Images Combining Self-Attention Multiple Instance Learning with a Recurrent Neural Networkcited by 1
- 2025Beyond Interpretability: When, Why, and How Sparse Autoencoders Enable Label-Free Visual Steeringcited by 1
Built on by
Not cited yet.