Sparse Autoencoders

Also known as SAE

Ask about this topic

Overview

Sparse autoencoders learn a dictionary of features that can reconstruct a model’s dense activations using only a few active features at a time, helping researchers investigate superposition in language, vision, and multimodal models. Common approaches include L1- and Top-k-based training, BatchTopK, and structured variants that model feature groups, hierarchies, or relationships across layers and modalities. Researchers use the resulting features to explain model behavior and to steer, audit, or remove targeted behaviors, while testing them with reconstruction and interpretability benchmarks as well as causal interventions. A central open tension is that sparse, human-readable features are not necessarily stable across training runs or faithful causes of behavior, and better reconstruction does not by itself guarantee reliable explanations or control.

Latest papers

CardsList

220 papers42 papers in the last four weeks, up 133% on the four weeks before. 0.2% of all new papers.

  1. Sparsity as a Key: Unlocking New Insights from Latent Structures for Out-of-Distribution Detection

    Apr 29, 2026Ahyoung Oh, Wonseok Shin, Songkuk KimSparse AutoencodersOOD Detection

    School of Integrated Technology, BK21 Graduate Program in Intelligent Semiconductor Technology Yonsei University

  2. A Unifying Framework for Unsupervised Concept Extraction

    Apr 27, 2026Chandler Squires, Pradeep RavikumarUnsupervised LearningParameter Identifiability

    Machine Learning Department, CMU

  3. Domain-Filtered Knowledge Graphs from Sparse Autoencoder Features

    Apr 26, 2026John Winnicki, Abeynaya Gnanasekaran, Eric DarveKG ConstructionLLM Interpretability

    Institute for Computational and Mathematical Engineering, Stanford University, Stanford, CA, USA

  4. From Concept-Aligned Tokens to Vulnerable Features: Mechanistic Localization of Jailbreaks

    Apr 25, 2026Nilanjana Das, Mathew Dawit, Aman Chadha +1LLM InterpretabilitySparse Autoencoders

    UMBC, Maryland, U.S.A. · Apple, California, U.S.A.

  5. From Tokens to Concepts: Leveraging SAE for SPLADE

    Apr 23, 2026Yuxuan Zong, Mathias Vast, Basile Van Cooten +2Learned Sparse RetrievalSparse Autoencoders

    Sorbonne Université, CNRS, ISIR · Sinequa by ChapsVision · CNRS, Sorbonne Université, ISIR

  6. Towards Understanding the Robustness of Sparse Autoencoders

    Apr 20, 2026Ahson Saiyed, Sabrina Sadiekh, Chirag AgarwalAdversarial Attacks on LLMsJailbreak Robustness

    University of Virginia · Independent Researcher

  7. Structural Instability of Feature Composition

    Apr 18, 2026Yunpeng ZhouFeature SuperpositionTransformer Interpretability

    Whiteknights House, Whiteknights, Reading, RG6 6UR

  8. Improving Sparse Autoencoder with Dynamic Attention

    Apr 16, 2026Dongsheng Wang, Jinsen Zhang, Dawei Su +1Sparse AutoencodersNeural Network Interpretability

    College of Computer Science and Software Engineering, Shenzhen University, China

  9. Sparse Autoencoder Decomposition of Clinical Sequence Model Representations: Feature Complexity, Task Specialisation, and Mortality Prediction

    Apr 13, 2026Chris Sainsbury, Feng Dong, Andreas KarwathEHR Foundation ModelsSparse Autoencoders

    School of Cardiovascular & Metabolic Health, University of Glasgow, UK · School of Medicine, University of Dundee, UK · NHS Greater Glasgow and Clyde, UK +2

  10. Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs

    Apr 10, 2026Jinqi Luo, Jinyu Yang, Tal Neiman +5MLLM SafetyMultimodal Large Language Models

    University of Pennsylvania · Amazon · University of Central Florida

  11. Improving Robustness In Sparse Autoencoders via Masked Regularization

    Apr 7, 2026Vivek Narayanaswamy, Kowshik Thopalli, Bhavya Kailkhura +1Sparse AutoencodersMechanistic Interpretability

    Lawrence Livermore National Laboratory, CA, USA

  12. Beyond Semantics: Disentangling Information Scope in Sparse Autoencoders for CLIP

    Apr 7, 2026Yusung Ro, Jaehyun Choi, Junmo KimNeural Network InterpretabilityVLM Interpretability

    KAIST · Korea AI Safety Institute, ETRI

  13. Interpreting Video Representations with Spatio-Temporal Sparse Autoencoders

    Apr 5, 2026Atahan Dokme, Sriram VishwanathSparse AutoencodersVideo Representation Learning

    Georgia Institute of Technology, Atlanta, Georgia, USA

  14. MetaSAEs: Joint Training with a Decomposability Penalty Produces More Atomic Sparse Autoencoder Latents

    Apr 3, 2026Matthew LevinsonSparse AutoencodersTransformer Interpretability

    Independent Researcher

  15. Stop Probing, Start Coding: Why Linear Probes and Sparse Autoencoders Fail at Compositional Generalisation

    Mar 30, 2026Vitória Barin-Pacela, Shruti Joshi, Isabela Camacho +2Sparse AutoencodersSparse Coding

    Mila - Québec AI Institute & Université de Montréal · Santa Clara University · Cold Spring Harbor Laboratory

  16. Sparse Visual Thought Circuits in Vision-Language Models

    Mar 26, 2026Yunpeng ZhouVLM InterpretabilitySparse Autoencoders

    University of Reading, Reading, UK

  17. Sparse Autoencoders for Interpretable Medical Image Representation Learning

    Mar 24, 2026Philipp Wesp, Robbie Holland, Vasiliki Sideri-Lampretsa +1Medical Image RetrievalExplainable Medical Image Analysis

    Stanford Center for Artificial Intelligence in Medicine and Imaging, Stanford University, Stanford, CA, USA · Department of Radiology, Stanford University, Stanford, CA, USA · Chair of AI in Healthcare and Medicine, Technical University of Munich, Munich, Germany +1

  18. DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment

    Mar 23, 2026James Wedgwood, Aashiq Muhamed, Mona T. Diab +1Preference AlignmentInference-Time Language Model Alignment

    Carnegie Mellon University

  19. Sparse Autoencoders Reveal Interpretable and Steerable Features in VLA Models

    Mar 19, 2026Aiden Swann, Lachlain McGranahan, Hugo Buurmeijer +2Vision-Language-Action ModelsInstruction Generalization in VLA Models

    Department of Mechanical Engineering, Stanford University · Department of Aeronautics & Astronautics, Stanford University · Department of Computer Science, Stanford University

  20. Towards Interpretable Framework for Neural Audio Codecs via Sparse Autoencoders: A Case Study on Accent Information

    Mar 18, 2026Shih-Heng Wang, Tiantian Feng, Aditya Kommineni +4Neural Audio CodecsSparse Autoencoders

    University of Southern California, USA

  21. Fundamental Limits of Neural Network Sparsification: Evidence from Catastrophic Interpretability Collapse

    Mar 18, 2026Dip Roy, Rajiv Misra, Sanjay Kumar SinghActivation SparsityNeural Network Interpretability

    Department of Computer Science and Engineering, Indian Institute of Technology, Patna, India · Department of Computer Science, Rajarshi School of Management Technology, Varanasi, India

  22. Dissecting Chronos: Sparse Autoencoders Reveal Causal Feature Hierarchies in Time Series Foundation Models

    Mar 10, 2026Anurag MishraTime-Series Foundation ModelsTransformer Interpretability

    Rochester Institute of Technology

  23. One Language, Two Scripts: Probing Script-Invariance in LLM Concept Representations

    Mar 9, 2026Sripad KarneLLM InterpretabilitySparse Autoencoders

    Columbia University

  24. Sparse Crosscoders for diffing MoEs and Dense models

    Mar 6, 2026Marmik Chaudhari, Nishkal Hundia, Idhant GulatiMixture-of-Experts InterpretabilityMixture-of-Experts Models

    University of California, Berkeley · University of Maryland

  25. Stable and Steerable Sparse Autoencoders with Weight Regularization

    Mar 4, 2026Piotr Jedryszek, Oliver M. CrookSparse AutoencodersLLM Interpretability

    Department of Biology, University of Oxford, Oxford, UK · Kavli Institute for Nanoscience Discovery, University of Oxford, Oxford, UK · Department of Chemistry, University of Oxford, Oxford, UK

  26. Step-Level Sparse Autoencoder for Reasoning Process Interpretation

    Mar 3, 2026Xuan Yang, Jiayu Liu, Yuhang Lai +3LLM InterpretabilitySparse Autoencoders

    Department of Data Science, City University of Hong Kong · Hong Kong Institute of AI for Science, City University of Hong Kong · Li Auto Inc +1

  27. Learning Retrieval Models with Sparse Autoencoders

    Feb 27, 2026Thibault Formal, Maxime Louis, Hervé Dejean +1Learned Sparse RetrievalMultilingual IR

    NAVER LABS Europe

  28. SynthSAEBench: Evaluating Sparse Autoencoders on Scalable Realistic Synthetic Data

    Feb 16, 2026David Chanin, Adrià Garriga-AlonsoSparse AutoencodersSynthetic Benchmark Generation

    University College London · 2MATS · 3Decode Research

  29. Sparse Autoencoders are Capable LLM Jailbreak Mitigators

    Feb 12, 2026Yannick Assogba, Jacopo Cortellazzi, Javier Abad +3Sparse AutoencodersLLM Jailbreak Attacks

    1Apple · 2ETH Zurich

  30. Variational Sparse Paired Autoencoders (vsPAIR) for Inverse Problems and Uncertainty Quantification

    Feb 3, 2026Jack Michael Solomon, Rishi Leburu, Matthias ChungVariational AutoencodersUncertainty Quantification

    Emory University Department of Mathematics