paper
referenced-only
2025
paper:fel-archetypal-sae-adaptive-and-stable-dicti-2025Archetypal sae: Adaptive and stable dictionary learning for concept extraction in large vision models
ByThomas Fel·Ekdeep Singh Lubana·Jacob S Prince·Matthew Kowal·Victor Boutin·Isabel Papadimitriou+4 more
Related work— refs + corpus + external arXiv
Cited / in-corpus / arXiv badges show which signals surfaced each row. Multi-source rows weighted higher.
- Sparse autoencoders reveal selective remapping of visual concepts during adaptationJinho Choi, Jaegul Choo, Steffen Schneider Hyesu Lim2025≈ 78%
- VL-SAE: Interpreting and Enhancing Vision-Language Alignment with a Unified Concept SetJunshu Sun, Qingming Huang, Shuhui Wang Shufan Shen2025≈ 78%
- Identifying Functionally Important Features with End-to-End Sparse Dictionary LearningJordan Taylor, Nicholas Goldowsky-Dill, Lee Sharkey Dan Braun2024≈ 77%
- LUCID-SAE: Learning Unified Vision-Language Sparse Codes for Interpretable Concept DiscoveryYunhe Gao, Gerasimos Chatzoudis, Zihan Dong, Guoning Zhang, Bangwei Guo, Yang Zhou, Mu Zhou, Dimitris Metaxas Difei Gu2026≈ 77%
- SAE-V: Interpreting Multimodal Models for Enhanced AlignmentChangye Li, Jiaming Ji, Yaodong Yang Hantao Lou2025≈ 76%
- Emergence and Evolution of Interpretable Concepts in Diffusion ModelsZalan Fabian, Mahdi Soltanolkotabi Berk Tinaz2026≈ 76%
- ≈ 75%
- Mechanistic Interpretability of Antibody Language Models Using SAEsOliver M. Turnbull, Anisha Parsan, Nithin Parsan, John J. Yang, Anna L. Beukenhorst, Charlotte M. Deane Rebonto Haque2026≈ 75%
- An X-Ray Is Worth 15 Features: Sparse Autoencoders for Interpretable Radiology Report GenerationHugo Fry, Nina Monta\~na-Brown, Ayodeji Ijishakin, Jack Gao, Stephanie Hyland, Daniel C. Alexander, Daniel C. Castro Ahmed Abdulaal2024≈ 75%
- A Unified Theory of Sparse Dictionary Learning in Mechanistic Interpretability: Piecewise Biconvexity and Spurious MinimaHarshvardhan Saini, Zhaoqian Yao, Zheng Lin, Yizhen Liao, Jingyi Cui, Yisen Wang, Mengnan Du, Dianbo Liu Yiming Tang2026≈ 75%
- ≈ 75%
- Position: Mechanistic Interpretability Should Prioritize Feature Consistency in SAEsAashiq Muhamed, Yujia Zheng, Lingjing Kong, Zeyu Tang, Mona T. Diab, Virginia Smith, Kun Zhang Xiangchen Song2025≈ 75%
- Mechanistic Knobs in LLMs: Retrieving and Steering High-Order Semantic Features via Sparse Autoencodersin corpus2026≈ 75%
- Sparse Autoencoder Features for Classifications and TransferabilityShan Chen, Kuleen Sasse, Hugo Aerts, Thomas Hartvigsen, Danielle S. Bitterman Jack Gallifant2026≈ 75%
- Group Equivariance Meets Mechanistic Interpretability: Equivariant Sparse AutoencodersAna Lucic Ege Erdogan2025≈ 75%
- Towards scientific discovery with dictionary learning: Extracting biological concepts from microscopy foundation modelsKristina Ulicna, Gemma Elyse Moran, Aditya Ravuri, Kian Kenyon-Dean, Cian Eastwood and Jason Hartford Konstantin Donhauser2025≈ 75%
- Interpreting Language Model Parametersin corpus2026≈ 74%
- ≈ 74%
- SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretabilityin corpus2025≈ 73%
- ≈ 73%
- Open Problems in Mechanistic Interpretabilityin corpus2025≈ 72%
- ≈ 72%
- ≈ 71%
- ≈ 71%
- ≈ 71%
- ≈ 70%
- ≈ 70%
- Active Inference, Curiosity and Insightin corpus2017≈ 70%
Similar preprints — Semantic Scholar
Cited by (3)
- Understanding sparse autoencoder scaling in the presence of feature manifolds
Sparse autoencoder scaling laws can break down when activations contain feature manifolds: if per-feature loss falls slowly with latent count, an SAE spends its latents tiling common manifolds instead
- Do Sparse Autoencoders Capture Concept Manifolds?
Sparse autoencoders trained on Llama3.1-8B (layer 19, 500M Pile tokens, 25 SAEs across Standard, JumpReLU, TopK, BatchTopK and Matryoshka architectures at VE > 0.85) do not capture continuous concept
- Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior
Manifold steering — intervening on model activations along paths constrained to lie on a learned activation manifold M_h rather than along Euclidean linear directions — produces behavioral trajectorie