mlsanityData to 7 Oct 2026

Beyond Input Activations: Identifying Influential Latents by Gradient Sparse Autoencoders

arXiv 2505.08080 ↗First posted 12 May 2025cs.LG cs.AI cs.CLcited by 3

Not on the map. Why?

Abstract (arXiv)

Sparse Autoencoders (SAEs) have recently emerged as powerful tools for interpreting and steering the internal representations of large language models (LLMs). However, conventional approaches to analyzing SAEs typically rely solely on input-side activations, without considering the causal influence between each latent feature and the model's output. This work is built on two key hypotheses: (1) activated latents do not contribute equally to the construction of the model's output, and (2) only latents with high causal influence are effective for model steering. To validate these hypotheses, we propose Gradient Sparse Autoencoder (GradSAE), a simple yet effective method that identifies the most influential latents by incorporating output-side gradient information.

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 19 of its 23 references that are in the map. First posted on arXiv 12 May 2025.

Builds on

19 of 23 references.

References · 19

  1. 2016SQuAD: 100,000+ Questions for Machine Comprehension of Textcited by 4,115
  2. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,359
  3. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 381
  4. 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoderscited by 283
  5. 2024Improving Dictionary Learning with Gated Sparse Autoencoderscited by 171
  6. 2016Linear Algebraic Structure of Word Senses, with Applications to Polysemycited by 165
  7. 2018Pathologies of Neural Models Make Interpretations Difficultcited by 140
  8. 2024Improving Steering Vectors by Targeting Sparse Autoencoder Featurescited by 86
  9. 2024Identifying Functionally Important Features with End-to-End Sparse Dictionary Learningcited by 64
  10. 2024Steering Knowledge Selection Behaviours in LLMs via SAE-Based Representation Engineeringcited by 60
  11. 2023From Language Modeling to Instruction Following: Understanding the Behavior Shift in LLMs after Instruction Tuningcited by 56
  12. 2025I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoderscited by 43
7 more
  1. 2025Interpretable Steering of Large Language Models with Feature Guided Activation Additionscited by 32
  2. 2025SAIF: A Sparse Autoencoder Framework for Interpreting and Steering Instruction Following of Language Modelscited by 32
  3. 2024Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencoderscited by 25
  4. 2025Interpreting and Steering LLMs with Mutual Information-based Explanations on Sparse Autoencoderscited by 21
  5. 2024Group-SAE: Efficient Training of Sparse Autoencoders for Large Language Models via Layer Groupscited by 10
  6. 2024Adaptive Sparse Allocation with Mutual Choice & Feature Choice Sparse Autoencoderscited by 9
  7. 2025Sparse Shift Autoencoders for Identifying Concepts from Large Language Model Activationscited by 9

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 1; 2026: 2.

3 papers.

  1. 2025ConceptViz: A Visual Analytics Approach for Exploring Concepts in Large Language Modelscited by 1
  2. 2026Gradient Atoms: Unsupervised Discovery, Attribution and Steering of Model Behaviors via Sparse Decomposition of Training Gradientscited by 0
  3. 2026SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?cited by 0