mlsanityData to 14 Oct 2026

Counterfactual Explanations on Robust Perceptual Geodesics

arXiv 2601.18678 ↗First posted 26 Jan 2026cs.LG cs.CV cs.HCcited by 2

Not on the map. Why?

Abstract (arXiv)

Latent-space optimization methods for counterfactual explanations - framed as minimal semantic perturbations that change model predictions - inherit the ambiguity of Wachter et al.'s objective: the choice of distance metric dictates whether perturbations are meaningful or adversarial. Existing approaches adopt flat or misaligned geometries, leading to off-manifold artifacts, semantic drift, or adversarial collapse. We introduce Perceptual Counterfactual Geodesics (PCG), a method that constructs counterfactuals by tracing geodesics under a perceptually Riemannian metric induced from robust vision features. This geometry aligns with human perception and penalizes brittle directions, enabling smooth, on-manifold, semantically valid transitions. Experiments on three vision datasets show that PCG outperforms baselines and reveals failure modes hidden under standard metrics.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 58 of its 77 references that are in the map. First posted on arXiv 26 Jan 2026.

Builds on

58 of 77 references.

References · 58

  1. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  2. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,364
  3. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,177
  4. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
  5. 2017A Unified Approach to Interpreting Model Predictionscited by 6,555
  6. 2018A Style-Based Generator Architecture for Generative Adversarial Networkscited by 6,013
  7. 2022Generative Adversarial Networkscited by 5,627
  8. 2014Deep Learning Face Attributes in the Wildcited by 3,693
  9. 2019Analyzing and Improving the Image Quality of StyleGANcited by 3,588
  10. 2017Axiomatic Attribution for Deep Networkscited by 3,384
  11. 2017Progressive Growing of GANs for Improved Quality, Stability, and Variationcited by 2,988
  12. 2013Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Mapscited by 2,371
38 more
  1. 2020Training Generative Adversarial Networks with Limited Datacited by 1,186
  2. 2021Alias-Free Generative Adversarial Networkscited by 1,184
  3. 2017SmoothGrad: removing noise by adding noisecited by 1,109
  4. 2019StarGAN v2: Diverse Image Synthesis for Multiple Domainscited by 1,098
  5. 2017Counterfactual Explanations without Opening the Black Box: Automated Decisions and the GDPRcited by 1,089
  6. 2018Robustness May Be at Odds with Accuracycited by 880
  7. 2018Actionable Recourse in Linear Classificationcited by 328
  8. 2017Generating Natural Adversarial Examplescited by 183
  9. 2017Latent Space Oddity: on the Curvature of Deep Generative Modelscited by 174
  10. 2018Disentangling Adversarial Robustness and Generalizationcited by 172
  11. 2015An open access repository of images on plant health to enable the development of mobile disease diagnosticscited by 135
  12. 2018Constructing Unrestricted Adversarial Examples with Generative Modelscited by 135
  13. 2019Towards Realistic Individual Recourse and Actionable Explanations in Black-Box Decision Making Systemscited by 131
  14. 2021Explaining in Style: Training a GAN to explain a classifier in StyleSpacecited by 122
  15. 2021Guided Integrated Gradients: An Adaptive Path Method for Removing Noisecited by 102
  16. 2022Diffusion Visual Counterfactual Explanationscited by 96
  17. 2017The Riemannian Geometry of Deep Generative Modelscited by 82
  18. 2019On the Connection Between Adversarial Robustness and Saliency Map Interpretabilitycited by 80
  19. 2019Explanation by Progressive Exaggerationcited by 76
  20. 2019Interpreting Adversarially Trained Convolutional Neural Networkscited by 75
  21. 2020Adversarial Robustness on In- and Out-Distribution Improves Explainabilitycited by 73
  22. 2022A Light Recipe to Train Robust Vision Transformerscited by 64
  23. 2021Do Input Gradients Highlight Discriminative Features?cited by 52
  24. 2021Exploring Counterfactual Explanations Through the Lens of Adversarial Examples: A Theoretical and Empirical Analysiscited by 52
  25. 2020GANterfactual - Counterfactual Explanations for Medical Non-Experts using Generative Adversarial Learningcited by 50
  26. 2019E-LPIPS: Robust Perceptual Image Similarity via Random Transformation Ensemblescited by 45
  27. 2020The Intriguing Relation Between Counterfactual Explanations and Adversarial Examplescited by 44
  28. 2019Are Perceptually-Aligned Gradients a General Property of Robust Classifiers?cited by 43
  29. 2023Fast Diffusion-Based Counterfactuals for Shortcut Removal and Generationcited by 41
  30. 2019On the computation of counterfactual explanations -- A surveycited by 33
  31. 2022Cycle-Consistent Counterfactuals by Latent Transformationscited by 33
  32. 2022Diffeomorphic Counterfactuals with Generative Modelscited by 29
  33. 2021Conditional Generative Models for Counterfactual Explanationscited by 28
  34. 2023R-LPIPS: An Adversarially Robust Perceptual Similarity Metriccited by 28
  35. 2020Semantics and explanation: why counterfactual explanations produce adversarial examples in deep neural networkscited by 24
  36. 2024Manifold Integrated Gradients: Riemannian Geometry for Feature Attributioncited by 21
  37. 2023Which Models have Perceptually-Aligned Gradients? An Explanation via Off-Manifold Robustnesscited by 20
  38. 2022Do Perceptually Aligned Gradients Imply Adversarial Robustness?cited by 15

50 most cited of 58 All 58 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Auditing Sybil: Explaining Deep Lung Cancer Risk Prediction Through Generative Interventional Attributionscited by 2
  2. 2026FCA-Guided Counterfactual Explanations for Multi-Modal Breast Cancer Diagnosis: A Framework Achieving Perfect Validity with Emergent Sparsitycited by 0