mlsanityData to 14 Oct 2026

Exploring Task Performance with Interpretable Models via Sparse Auto-Encoders

arXiv 2507.06427 ↗First posted 8 Jul 2025cs.CL cs.LGcited by 1

Not on the map. Why?

Abstract (arXiv)

Large Language Models (LLMs) are traditionally viewed as black-box algorithms, therefore reducing trustworthiness and obscuring potential approaches to increasing performance on downstream tasks. In this work, we apply an effective LLM decomposition method using a dictionary-learning approach with sparse autoencoders. This helps extract monosemantic features from polysemantic LLM neurons. Remarkably, our work identifies model-internal misunderstanding, allowing the automatic reformulation of the prompts with additional annotations to improve the interpretation by LLMs. Moreover, this approach demonstrates a significant performance improvement in downstream tasks, such as mathematical reasoning and metaphor detection.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 28 of its 45 references that are in the map. First posted on arXiv 8 Jul 2025.

Builds on

28 of 45 references.

References · 28

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
  2. 2020Language Models are Few-Shot Learnerscited by 32,419
  3. 2024The Llama 3 Herd of Modelscited by 14,715
  4. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  5. 2024Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phonecited by 1,917
  6. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  7. 2017Towards A Rigorous Science of Interpretable Machine Learningcited by 1,416
  8. 2022Solving Quantitative Reasoning Problems with Language Modelscited by 1,392
  9. 2022Toy Models of Superpositioncited by 971
  10. 2018What you can cram into a single vector: Probing sentence embeddings for linguistic propertiescited by 457
  11. 2018Analysis Methods in Neural Language Processing: A Surveycited by 277
  12. 2016Linear Algebraic Structure of Word Senses, with Applications to Polysemycited by 165
16 more
  1. 2015Visualizing and Understanding Recurrent Networkscited by 160
  2. 2024OpenMathInstruct-1: A 1.8 Million Math Instruction Tuning Datasetcited by 143
  3. 2023Large language models in healthcare and medical domain: A reviewcited by 127
  4. 2021Transformer visualization via dictionary learning: contextualized embedding as a linear superposition of transformer factorscited by 108
  5. 2018Beyond Word Importance: Contextual Decomposition to Extract Interactions from LSTMscited by 67
  6. 2022Post-Hoc Explanations Fail to Achieve their Purpose in Adversarial Contextscited by 53
  7. 2021MelBERT: Metaphor Detection via Contextualized Late Interaction using Metaphorical Identification Theoriescited by 46
  8. 2015Sparse Overcomplete Word Vector Representationscited by 44
  9. 2023Retrieval-augmented Generation to Improve Math Question-Answering: Trade-offs Between Groundedness and Human Preferencecited by 38
  10. 2024LawLLM: Law Large Language Model for the US Legal Systemcited by 37
  11. 2024What Makes Math Word Problems Challenging for LLMs?cited by 19
  12. 2023FrameBERT: Conceptual Metaphor Detection with Frame Embedding Learningcited by 18
  13. 2024BLADE: Enhancing Black-box Large Language Models with Small Domain-Specific Modelscited by 15
  14. 2024Everything Everywhere All at Once: LLMs can In-Context Learn Multiple Tasks in Superpositioncited by 14
  15. 2023Metaphor Detection with Effective Context Denoisingcited by 13
  16. 2018LISA: Explaining Recurrent Neural Network Judgments via Layer-wIse Semantic Accumulation and Example to Pattern Transformationcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026The Granularity Axis: A Micro-to-Macro Latent Direction for Social Roles in Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Models
  2. 2025Interpreting and Steering LLMs with Mutual Information-based Explanations on Sparse Autoencoders
  3. 2025Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words
  4. 2025I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders
  5. 2025Sparse Autoencoders Can Capture Language-Specific Concepts Across Diverse Languages
  6. 2024Automatically Interpreting Millions of Features in Large Language Models
  7. 2025Sparse Shift Autoencoders for Identifying Concepts from Large Language Model Activations
  8. 2026Sparse Autoencoders for Mechanistic Interpretability in NLP: A Critical Survey of Recoverability, Interpretability, and Utility