mlsanityData to 14 Oct 2026

LEACE: Perfect linear concept erasure in closed form

arXiv 2306.03819 ↗First posted 6 Jun 2023cs.LG cs.CL cs.CYcited by 267

Not on the map. Why?

Cited by 52 in LLM safety and security, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Concept erasure aims to remove specified features from an embedding. It can improve fairness (e.g. preventing a classifier from using gender or race) and interpretability (e.g. removing a concept to observe changes in model behavior). We introduce LEAst-squares Concept Erasure (LEACE), a closed-form method which provably prevents all linear classifiers from detecting a concept while changing the embedding as little as possible, as measured by a broad class of norms. We apply LEACE to large language models with a novel procedure called "concept scrubbing," which erases target concept information from every layer in the network. We demonstrate our method on two tasks: measuring the reliance of language models on part-of-speech information, and reducing gender bias in BERT embeddings. Code is available at https://github.com/EleutherAI/concept-erasure.

Key facts. Cited by 267 papers in the map (data through 14 Oct 2026). Builds on 35 of its 57 references that are in the map. First posted on arXiv 6 Jun 2023.

Builds on

35 of 57 references.

References · 35

  1. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 14,132
  2. 2020The Pile: An 800GB Dataset of Diverse Text for Language Modelingcited by 2,411
  3. 2023Pythia: A Suite for Analyzing Large Language Models Across Training and Scalingcited by 1,786
  4. 2016Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddingscited by 1,312
  5. 2019BERT Rediscovers the Classical NLP Pipelinecited by 1,059
  6. 2017Counterfactual Fairnesscited by 804
  7. 2018Mitigating Unwanted Biases with Adversarial Learningcited by 663
  8. 2023Eliciting Latent Predictions from Transformers with the Tuned Lenscited by 623
  9. 2020Null It Out: Guarding Protected Attributes by Iterative Nullspace Projectioncited by 455
  10. 2019Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Settingcited by 349
  11. 2020Universal Dependencies v2: An Evergrowing Multilingual Treebank Collectioncited by 271
  12. 2015Censoring Representations with an Adversarycited by 192
23 more
  1. 2018Adversarial Removal of Demographic Attributes from Text Datacited by 170
  2. 2020A Theory of Usable Information Under Computational Constraintscited by 162
  3. 2021All Bark and No Bite: Rogue Dimensions in Transformer Language Models Obscure Representational Qualitycited by 156
  4. 2017Controllable Invariance through Adversarial Feature Learningcited by 127
  5. 2021Inducing Causal Structure for Interpretable Neural Networkscited by 110
  6. 2023Interpretability at Scale: Identifying Causal Mechanisms in Alpacacited by 103
  7. 2022Linear Adversarial Concept Erasurecited by 101
  8. 2021Counterfactual Interventions Reveal the Causal Effect of Relative Clause Representations on Agreement Predictioncited by 59
  9. 2016Adversarial Deep Averaging Networks for Cross-Lingual Sentiment Classificationcited by 48
  10. 2022Kernelized Concept Erasurecited by 48
  11. 2022Probing Classifiers are Unreliable for Concept Removal and Detectioncited by 41
  12. 2020OSCaR: Orthogonal Subspace Correction and Rectification of Biases in Word Embeddingscited by 40
  13. 2021The Low-Dimensional Linear Geometry of Contextualized Word Representationscited by 40
  14. 2022Causal Conceptions of Fairness and their Consequencescited by 38
  15. 2022Gold Doesn't Always Glitter: Spectral Removal of Linear and Nonlinear Guarded Attribute Informationcited by 34
  16. 2022Can Transformer be Too Compositional? Analysing Idiom Processing in Neural Machine Translationcited by 28
  17. 2023Efficient fair PCA for fair representation learningcited by 23
  18. 2022Better Hit the Nail on the Head than Beat around the Bush: Removing Protected Attributes with a Single Projectioncited by 17
  19. 2019On the Global Optima of Kernelized Adversarial Representation Learningcited by 14
  20. 2021The Rediscovery Hypothesis: Language Models Need to Meet Linguisticscited by 7
  21. 2023Erasure of Unaligned Attributes from Neural Representationscited by 6
  22. 2021Obstructing Classification via Projectioncited by 5
  23. 2022Log-linear Guardedness and its Implicationscited by 1

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 20; 2024: 60; 2025: 57; 2026: 130.

267 papers.

Sort:

By area:

  1. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  2. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  3. 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 712
  4. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 509
  5. 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 435
  6. 2024MUSE: Machine Unlearning Six-Way Evaluation for Language Modelscited by 261
  7. 2023Large Language Model Alignment: A Surveycited by 243
  8. 2023In-Context Unlearning: Language Models as Few Shot Unlearnerscited by 224
  9. 2023An Overview of Catastrophic AI Riskscited by 222
  10. 2024ReFT: Representation Finetuning for Language Modelscited by 197
  11. 2023Knowledge Editing for Large Language Models: A Surveycited by 194
  12. 2026Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnetcited by 191
38 more
  1. 2024A Comprehensive Study of Knowledge Editing for Large Language Modelscited by 163
  2. 2024Improving Instruction-Following in Language Models through Activation Steeringcited by 160
  3. 2023Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against Extraction Attackscited by 158
  4. 2024Tamper-Resistant Safeguards for Open-Weight LLMscited by 138
  5. 2024Large Language Model Unlearning via Embedding-Corrupted Promptscited by 116
  6. 2025The Geometry of Refusal in Large Language Models: Concept Cones and Representational Independencecited by 115
  7. 2024AtP*: An efficient and scalable method for localizing LLM behaviour to componentscited by 111
  8. 2024Guardrail Baselines for Unlearning in LLMscited by 96
  9. 2023Stay on topic with Classifier-Free Guidancecited by 91
  10. 2024Detoxifying Large Language Models via Knowledge Editingcited by 87
  11. 2024Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Controlcited by 86
  12. 2024A Primer on the Inner Workings of Transformer-based Language Modelscited by 85
  13. 2024RAVEL: Evaluating Interpretability Methods on Disentangling Language Model Representationscited by 79
  14. 2024RWKU: Benchmarking Real-World Knowledge Unlearning for Large Language Modelscited by 78
  15. 2024Adaptive Activation Steering: A Tuning-Free LLM Truthfulness Improvement Method for Diverse Hallucinations Categoriescited by 77
  16. 2024LLM Unlearning via Loss Adjustment with Only Forget Datacited by 75
  17. 2025Convergent Linear Representations of Emergent Misalignmentcited by 71
  18. 2023Improving Activation Steering in Language Models with Mean-Centringcited by 68
  19. 2023The Curious Case of Hallucinatory (Un)answerability: Finding Truths in the Hidden States of Over-Confident Large Language Modelscited by 55
  20. 2024Representation Surgery: Theory and Practice of Affine Steeringcited by 50
  21. 2024Recurrent Neural Networks Learn to Store and Generate Sequences using Non-Linear Representationscited by 47
  22. 2024CausalGym: Benchmarking causal interpretability methods on linguistic taskscited by 46
  23. 2025A Unified Understanding and Evaluation of Steering Methodscited by 46
  24. 2024Digital Forgetting in Large Language Models: A Survey of Unlearning Methodscited by 40
  25. 2025Steering Out-of-Distribution Generalization with Concept Ablation Fine-Tuningcited by 39
  26. 2024Machine Unlearning Fails to Remove Data Poisoning Attackscited by 37
  27. 2024Immunization against harmful fine-tuning attackscited by 35
  28. 2024In-Context Editing: Learning Knowledge from Self-Induced Distributionscited by 34
  29. 2024Gradient Routing: Masking Gradients to Localize Computation in Neural Networkscited by 34
  30. 2025Localized Concept Erasure for Text-to-Image Diffusion Models Using Training-Free Gated Low-Rank Adaptationcited by 33
  31. 2024Preference Tuning For Toxicity Mitigation Generalizes Across Languagescited by 31
  32. 2025To Steer or Not to Steer? Mechanistic Error Reduction with Abstention for Language Modelscited by 29
  33. 2023Emptying the Ocean with a Spoon: Should We Edit Models?cited by 27
  34. 2024Exploring Safety-Utility Trade-Offs in Personalized Language Modelscited by 27
  35. 2026Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behaviorcited by 26
  36. 2023A Geometric Notion of Causal Probingcited by 25
  37. 2024Optimal ablation for interpretabilitycited by 25
  38. 2023Debiasing Algorithm through Model Adaptationcited by 24

50 most cited of 267 All 267 (JSON)