mlsanityData to 7 Oct 2026

Exploiting the Relationship Between Kendall's Rank Correlation and Cosine Similarity for Attribution Protection

arXiv 2205.07279 ↗First posted 15 May 2022cs.LG cs.AIcited by 6

Not on the map. Why?

Abstract (arXiv)

Model attributions are important in deep neural networks as they aid practitioners in understanding the models, but recent studies reveal that attributions can be easily perturbed by adding imperceptible noise to the input. The non-differentiable Kendall's rank correlation is a key performance index for attribution protection. In this paper, we first show that the expected Kendall's rank correlation is positively correlated to cosine similarity and then indicate that the direction of attribution is the key to attribution robustness. Based on these findings, we explore the vector space of attribution to explain the shortcomings of attribution defense methods using ℓp norm and propose integrated gradient regularizer (IGR), which maximizes the cosine similarity between natural and perturbed attributions. Our analysis further exposes that IGR encourages neurons with the same activation states for natural samples and the corresponding perturbed samples, which is shown to induce robustness to gradient-based attribution methods. Our experiments on different models and datasets confirm our analysis on attribution protection and demonstrate a decent improvement in adversarial robustness.

Key facts. Cited by 6 papers in the map (data through 7 Oct 2026). Builds on 26 of its 42 references that are in the map. First posted on arXiv 15 May 2022.

Builds on

26 of 42 references.

References · 26

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  3. 2014Explaining and Harnessing Adversarial Examplescited by 7,287
  4. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
  5. 2013Intriguing properties of neural networkscited by 5,114
  6. 2017Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithmscited by 4,283
  7. 2017Axiomatic Attribution for Deep Networkscited by 3,350
  8. 2016Towards Evaluating the Robustness of Neural Networkscited by 2,976
  9. 2013Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Mapscited by 2,348
  10. 2013Visualizing and Understanding Convolutional Networkscited by 2,278
  11. 2019Theoretically Principled Trade-off between Robustness and Accuracycited by 1,510
  12. 2017Learning Important Features Through Propagating Activation Differencescited by 1,454
14 more
  1. 2016Adversarial Machine Learning at Scalecited by 934
  2. 2015The Limitations of Deep Learning in Adversarial Settingscited by 784
  3. 2015Distillation as a Defense to Adversarial Perturbations against Deep Neural Networkscited by 675
  4. 2017Interpretation of Neural Networks is Fragilecited by 443
  5. 2016European Union regulations on algorithmic decision-making and a "right to explanation"cited by 343
  6. 2019Explanations can be manipulated and geometry is to blamecited by 215
  7. 2017Visualizing Deep Neural Network Decisions: Prediction Difference Analysiscited by 166
  8. 2019Fooling Neural Network Interpretations via Adversarial Model Manipulationcited by 112
  9. 2020Explainable deep learning models in medical image analysiscited by 111
  10. 2019Robust Attribution Regularizationcited by 46
  11. 2020Proper Network Interpretability Helps Adversarial Robustness in Classificationcited by 41
  12. 2020Smoothed Geometry for Robust Attributioncited by 38
  13. 2020FAR: A General Framework for Attributional Robustnesscited by 13
  14. 2020Enhanced Regularizers for Attributional Robustnesscited by 12

Built on by

Cited by year

Citing papers per year: 2022: 2; 2023: 3; 2024: 1.

6 papers.

  1. 2022On the Robustness of Explanations of Deep Neural Network Models: A Surveycited by 5
  2. 2023Rethinking Robustness of Model Attributionscited by 3
  3. 2023A Practical Upper Bound for the Worst-Case Attribution Deviationscited by 2
  4. 2022Provable Robust Saliency-based Explanationscited by 1
  5. 2024Certified ℓ2 Attribution Robustness via Uniformly Smoothed Attributionscited by 1
  6. 2023Robust Ranking Explanationscited by 0