mlsanityData to 7 Oct 2026

Fairwashing Explanations with Off-Manifold Detergent

arXiv 2007.09969 ↗First posted 20 Jul 2020cs.LG stat.MLcited by 71

Not on the map. Why?

Abstract (arXiv)

Explanation methods promise to make black-box classifiers more transparent. As a result, it is hoped that they can act as proof for a sensible, fair and trustworthy decision-making process of the algorithm and thereby increase its acceptance by the end-users. In this paper, we show both theoretically and experimentally that these hopes are presently unfounded. Specifically, we show that, for any classifier g, one can always construct another classifier tildeg which has the same behavior on the data (same train, validation, and test error) but has arbitrarily manipulated explanation maps. We derive this statement theoretically using differential geometry and demonstrate it experimentally for various explanation methods, architectures, and datasets. Motivated by our theoretical insights, we then propose a modification of existing explanation methods which makes them significantly more robust.

Key facts. Cited by 71 papers in the map (data through 7 Oct 2026). First posted on arXiv 20 Jul 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 13; 2022: 19; 2023: 17; 2024: 9; 2025: 9; 2026: 4.

71 papers.

Sort:

By area:

  1. 2022From Anecdotal Evidence to Quantitative Evaluation Methods: A Systematic Review on Evaluating Explainable AIcited by 299
  2. 2023Manifold Preserving Guided Diffusioncited by 153
  3. 2022Bias Mitigation for Machine Learning Classifiers: A Comprehensive Surveycited by 125
  4. 2021Counterfactual Explanations Can Be Manipulatedcited by 113
  5. 2023Faithfulness Tests for Natural Language Explanationscited by 107
  6. 2022Post hoc Explanations may be Ineffective for Detecting Unknown Spurious Correlationcited by 59
  7. 2022Framework for Evaluating Faithfulness of Local Explanationscited by 55
  8. 2022Post-Hoc Explanations Fail to Achieve their Purpose in Adversarial Contextscited by 53
  9. 2022Rethinking Explainability as a Dialogue: A Practitioner's Perspectivecited by 52
  10. 2022Leveraging Explanations in Interactive Machine Learning: An Overviewcited by 52
  11. 2022Beyond Explaining: Opportunities and Challenges of XAI-Based Model Improvementcited by 48
  12. 2023Adversarial attacks and defenses in explainable artificial intelligence: A surveycited by 47
38 more
  1. 2021Evaluating Robustness of Counterfactual Explanationscited by 41
  2. 2021A Survey on the Robustness of Feature Importance and Counterfactual Explanationscited by 38
  3. 2023Don't trust your eyes: on the (un)reliability of feature visualizationscited by 32
  4. 2022Diffeomorphic Counterfactuals with Generative Modelscited by 29
  5. 2024Decoupling Pixel Flipping and Occlusion Strategy for Consistent XAI Benchmarkscited by 28
  6. 2021A Sociotechnical View of Algorithmic Fairnesscited by 24
  7. 2021Selective Ensembles for Consistent Predictionscited by 23
  8. 2022Robust Explanation Constraints for Neural Networkscited by 23
  9. 2021Characterizing the risk of fairwashingcited by 21
  10. 2022Challenges in Applying Explainability Methods to Improve the Fairness of NLP Modelscited by 20
  11. 2023Which Models have Perceptually-Aligned Gradients? An Explanation via Off-Manifold Robustnesscited by 20
  12. 2023On the Robustness of Removal-Based Feature Attributionscited by 18
  13. 2021On the Lack of Robust Interpretability of Neural Text Classifierscited by 17
  14. 2021PredDiff: Explanations and Interactions from Conditional Expectationscited by 16
  15. 2023A Critical Survey on Fairness Benefits of Explainable AIcited by 16
  16. 2022The Manifold Hypothesis for Gradient-Based Explanationscited by 15
  17. 2022On the Robustness of Bayesian Neural Networks to Adversarial Attackscited by 12
  18. 2021Attributing Fair Decisions with Attention Interventionscited by 10
  19. 2021Making Corgis Important for Honeycomb Classification: Adversarial Attacks on Concept-based Explainability Toolscited by 9
  20. 2023Goodhart's Law Applies to NLP's Explanation Benchmarkscited by 9
  21. 2022Interpretability Guarantees with Merlin-Arthur Classifierscited by 8
  22. 2022On the amplification of security and privacy risks by post-hoc explanations in machine learning modelscited by 8
  23. 2023Manifold Restricted Interventional Shapley Valuescited by 8
  24. 2024Rethinking Explainable Machine Learning as Applied Statisticscited by 8
  25. 2023Adversarial Attacks on the Interpretation of Neuron Activation Maximizationcited by 7
  26. 2021Understanding Prediction Discrepancies in Machine Learning Classifierscited by 6
  27. 2023Sum-of-Parts: Self-Attributing Neural Networks with End-to-End Learning of Feature Groupscited by 6
  28. 2025Robust ML Auditing using Prior Knowledgecited by 6
  29. 2023Explainable artificial intelligence approaches for brain-computer interfaces: a review and design spacecited by 5
  30. 2024When Can You Trust Your Explanations? A Robustness Analysis on Feature Importancescited by 5
  31. 2021Seven challenges for harmonizing explainability requirementscited by 4
  32. 2022Backdooring Explainable Machine Learningcited by 4
  33. 2022Towards More Robust Interpretation via Local Gradient Alignmentcited by 4
  34. 2023Survey on AI Ethics: A Socio-technical Perspectivecited by 4
  35. 2022Training Characteristic Functions with Reinforcement Learning: XAI-methods play Connect Fourcited by 3
  36. 2023On Minimizing the Impact of Dataset Shifts on Actionable Explanationscited by 3
  37. 2024Manipulating Feature Visualizations with Gradient Slingshotscited by 3
  38. 2025Birds look like cars: Adversarial analysis of intrinsically interpretable deep learningcited by 3

50 most cited of 71 All 71 (JSON)