mlsanityData to 7 Oct 2026

Kernelized Concept Erasure

arXiv 2201.12191 ↗First posted 28 Jan 2022cs.LG cs.CLcited by 47

Not on the map. Why?

Abstract (arXiv)

The representation space of neural models for textual data emerges in an unsupervised manner during training. Understanding how those representations encode human-interpretable concepts is a fundamental problem. One prominent approach for the identification of concepts in neural representations is searching for a linear subspace whose erasure prevents the prediction of the concept from the representations. However, while many linear erasure algorithms are tractable and interpretable, neural networks do not necessarily represent concepts in a linear manner. To identify non-linearly encoded concepts, we propose a kernelization of a linear minimax game for concept erasure. We demonstrate that it is possible to prevent specific non-linear adversaries from predicting the concept. However, the protection does not transfer to different nonlinear adversaries. Therefore, exhaustively erasing a non-linearly encoded concept remains an open problem.

Key facts. Cited by 47 papers in the map (data through 7 Oct 2026). Builds on 19 of its 51 references that are in the map. First posted on arXiv 28 Jan 2022.

Builds on

19 of 51 references.

References · 19

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  2. 2013Efficient Estimation of Word Representations in Vector Spacecited by 4,346
  3. 2021Convex optimizationcited by 2,479
  4. 2019ViLBERT: Pretraining Task-Agnostic Visiolinguistic Representations for Vision-and-Language Taskscited by 2,219
  5. 2017Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)cited by 1,309
  6. 2016Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddingscited by 1,304
  7. 2020Null It Out: Guarding Protected Attributes by Iterative Nullspace Projectioncited by 442
  8. 2019Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Settingcited by 346
  9. 2019Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Themcited by 268
  10. 2014SimLex-999: Evaluating Semantic Models with (Genuine) Similarity Estimationcited by 170
  11. 2019It's All in the Name: Mitigating Gender Bias with Name-Based Counterfactual Data Substitutioncited by 102
  12. 2020Multimodal Pretraining Unmasked: A Meta-Analysis and a Unified Framework of Vision-and-Language BERTscited by 71
7 more
  1. 2019Attenuating Bias in Word Vectorscited by 64
  2. 2021Counterfactual Interventions Reveal the Causal Effect of Relative Clause Representations on Agreement Predictioncited by 58
  3. 2021The Low-Dimensional Linear Geometry of Contextualized Word Representationscited by 39
  4. 2021A Non-Linear Structural Probecited by 31
  5. 2020It's not Greek to mBERT: Inducing Word-Level Translations from Multilingual BERTcited by 30
  6. 2020Exploring the Linear Subspace Hypothesis in Gender Bias Mitigationcited by 25
  7. 2019On the Global Optima of Kernelized Adversarial Representation Learningcited by 14

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 4; 2023: 13; 2024: 12; 2025: 7; 2026: 11.

47 papers.

Sort:

By area:

  1. 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 412
  2. 2023LEACE: Perfect linear concept erasure in closed formcited by 254
  3. 2023In-Context Unlearning: Language Models as Few Shot Unlearnerscited by 214
  4. 2023Emergent and Predictable Memorization in Large Language Modelscited by 147
  5. 2022Probing for the Usage of Grammatical Numbercited by 57
  6. 2022Self-Destructing Models: Increasing the Costs of Harmful Dual Uses of Foundation Modelscited by 56
  7. 2024Representation Surgery: Theory and Practice of Affine Steeringcited by 47
  8. 2022Probing Classifiers are Unreliable for Concept Removal and Detectioncited by 40
  9. 2024Machine Unlearning Fails to Remove Data Poisoning Attackscited by 35
  10. 2025Steering Out-of-Distribution Generalization with Concept Ablation Fine-Tuningcited by 34
  11. 2024Exploring Safety-Utility Trade-Offs in Personalized Language Modelscited by 26
  12. 2023Shielded Representations: Protecting Sensitive Attributes Through Iterative Gradient-Based Projectioncited by 25
35 more
  1. 2024The Quest for the Right Mediator: Surveying Mechanistic Interpretability Through the Lens of Causal Mediation Analysiscited by 24
  2. 2022Naturalistic Causal Probing for Morpho-Syntaxcited by 19
  3. 2024Gumbel Counterfactual Generation From Language Modelscited by 13
  4. 2023Robust Concept Erasure via Kernelized Rate-Distortion Maximizationcited by 10
  5. 2024Opt-Out: Investigating Entity-Level Unlearning for Large Language Models via Optimal Transportcited by 10
  6. 2024Protecting Privacy Through Approximating Optimal Parameters for Sequence Unlearning in Language Modelscited by 10
  7. 2024How Reliable are Causal Probing Interventions?cited by 9
  8. 2025Memory in Large Language Models: Mechanisms, Evaluation and Evolutioncited by 8
  9. 2025Can Prompts Rewind Time for LLMs? Evaluating the Effectiveness of Prompted Knowledge Cutoffscited by 8
  10. 2023Gen-Z: Generative Zero-Shot Text Classification with Contextualized Label Descriptionscited by 7
  11. 2023Removing Spurious Concepts from Neural Network Representations via Joint Subspace Estimationcited by 5
  12. 2023Fair Streaming Principal Component Analysis: Statistical and Algorithmic Viewpointcited by 5
  13. 2024Focus On This, Not That! Steering LLMs with Adaptive Feature Specificationcited by 5
  14. 2025Fundamental Limits of Perfect Concept Erasurecited by 5
  15. 2023Competence-Based Analysis of Language Modelscited by 4
  16. 2024Explaining Text Classifiers with Counterfactual Representationscited by 3
  17. 2024Leveraging Prototypical Representations for Mitigating Social Bias without Demographic Informationcited by 3
  18. 2026What Do EEG Foundation Models Capture from Human Brain Signals?cited by 3
  19. 2023Counterfactual Probing for the Influence of Affect and Specificity on Intergroup Biascited by 2
  20. 2023How To Build Competitive Multi-gender Speech Translation Models For Controlling Speaker Gender Translationcited by 2
  21. 2023Counterfactually Probing Language Identity in Multilingual Modelscited by 2
  22. 2025Improving Causal Interventions in Amnesic Probing with Mean Projection or LEACEcited by 2
  23. 2025Nonlinear Concept Erasure: a Density Matching Approachcited by 2
  24. 2026No More, No Less: Least-Privilege Language Modelscited by 1
  25. 2026Obliviator Reveals the Cost of Nonlinear Guardedness in Concept Erasurecited by 1
  26. 2023The Ethics of Automating Legal Actorscited by 0
  27. 2025Extending Fair Null-Space Projections for Continuous Attributes to Kernel Methodscited by 0
  28. 2026Refit the Probe: Single-Direction Ablation Is Not a Necessity Testcited by 0
  29. 2026Debiasing Without Protected Attributes: Latent Concept Erasure from Textual Profilescited by 0
  30. 2026MPSelectTune: Prompt-type Selection for Fine-tuning improves Concept Unlearning in LLMscited by 0
  31. 2026MANCE: Manifold Aware Concept Erasurecited by 0
  32. 2026SCOPE: Entanglement Frontier Escape for Source-Free Class Unlearningcited by 0
  33. 2026MUtE: A Dual Framework for Concept Erasure and Counterfactual Interventionscited by 0
  34. 2026Hidden not Deleted: How Networks Suppress Entangled Featurescited by 0
  35. 2026Is invariance all you need for algorithmic fairness? Removing demographic information can create new biascited by 0