mlsanityData to 7 Oct 2026

Concept-SAE: A Controllable and Invertible Concept Interface for Sparse Autoencoders

arXiv 2509.22015 ↗First posted 26 Sep 2025cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Standard Sparse Autoencoders (SAEs) excel at discovering a dictionary of a model's learned features, providing a powerful lens for passive feature discovery. However, this passive nature makes it difficult to systematically evaluate or analyze concepts that users explicitly care about. We introduce Concept-SAE, a framework that augments SAEs with a structured and controllable interface for probing user-defined concepts. Concept-SAE decomposes an activation subspace into two orthogonal components: Concept Tokens, which are aligned to externally specified semantics through dual supervision on both concept existence and spatial localization, and Free Tokens, which operate like standard SAEs to capture all remaining information. This hybrid disentanglement strategy ensures that Concept Tokens are faithful, spatially grounded, and cleanly separated from the residual subspace while preserving the ability of SAEs for open-ended concept discovery. We conduct extensive experiments demonstrating that Concept-SAE yields high-fidelity, well-localized, and strongly disentangled concept representations, outperforming alternatives in interface quality. Finally, we validate the utility of this conceptual interface through three diagnostic evaluations: a detection test on classifying adversarial image samples, a controllability test focusing on controlled counterfactual editing and a stability test using adversarial perturbations. Together, these results show that Concept-SAE equips SAEs with a reliable mechanism for evaluating, probing, and diagnosing user-defined concepts.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 31 of its 39 references that are in the map. First posted on arXiv 26 Sep 2025.

Builds on

31 of 39 references.

References · 31

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  3. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  4. 2014Explaining and Harnessing Adversarial Examplescited by 7,287
  5. 2014Deep Learning Face Attributes in the Wildcited by 3,679
  6. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,359
  7. 2020Concept Bottleneck Modelscited by 1,022
  8. 2024Scaling and evaluating sparse autoencoderscited by 530
  9. 2017Feature Squeezing: Detecting Adversarial Examples in Deep Neural Networkscited by 484
  10. 2017Mitigating Adversarial Effects Through Randomizationcited by 382
  11. 2019On Completeness-aware Concept-Based Explanations in Deep Neural Networkscited by 239
  12. 2024Improving Dictionary Learning with Gated Sparse Autoencoderscited by 171
19 more
  1. 2018Visual Interpretability for Deep Learning: a Surveycited by 155
  2. 2025Are Sparse Autoencoders Useful? A Case Study in Sparse Probingcited by 109
  3. 2024Discover-then-Name: Task-Agnostic Concept Bottlenecks via Automated Concept Discoverycited by 91
  4. 2025Sparse Autoencoders Trained on the Same Data Learn Different Featurescited by 86
  5. 2018Shield: Fast, Practical Defense and Vaccination for Deep Learning using JPEG Compressioncited by 74
  6. 2025A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Modelscited by 73
  7. 2018Deflecting Adversarial Attacks with Pixel Deflectioncited by 71
  8. 2024Sparse autoencoders reveal selective remapping of visual concepts during adaptationcited by 66
  9. 2022Overlooked factors in concept-based explanations: Dataset choice, concept learnability, and human capabilitycited by 46
  10. 2024Interpreting Attention Layer Outputs with Sparse Autoencoderscited by 43
  11. 2023Learning to Receive Help: Intervention-Aware Concept Embedding Modelscited by 41
  12. 2024Efficient Dictionary Learning with Switch Sparse Autoencoderscited by 38
  13. 2024The Missing Curve Detectors of InceptionV1: Applying Sparse Autoencoders to InceptionV1 Early Visioncited by 37
  14. 2024Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencoderscited by 25
  15. 2025Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Wordscited by 17
  16. 2024Sparse Autoencoders Enable Scalable and Reliable Circuit Identification in Language Modelscited by 13
  17. 2025Measuring and Guiding Monosemanticitycited by 10
  18. 2025Probing the Representational Power of Sparse Autoencoders in Vision Modelscited by 9
  19. 2024Multimodal Attack Detection for Action Recognition Modelscited by 5

Built on by

Not cited yet.