mlsanityData to 7 Oct 2026

Guided Integrated Gradients: An Adaptive Path Method for Removing Noise

arXiv 2106.09788 ↗First posted 17 Jun 2021cs.CV cs.LGcited by 102

Not on the map. Why?

Abstract (arXiv)

Integrated Gradients (IG) is a commonly used feature attribution method for deep neural networks. While IG has many desirable properties, the method often produces spurious/noisy pixel attributions in regions that are not related to the predicted class when applied to visual models. While this has been previously noted, most existing solutions are aimed at addressing the symptoms by explicitly reducing the noise in the resulting attributions. In this work, we show that one of the causes of the problem is the accumulation of noise along the IG path. To minimize the effect of this source of noise, we propose adapting the attribution path itself -- conditioning the path not just on the image but also on the model being explained. We introduce Adaptive Path Methods (APMs) as a generalization of path methods, and Guided IG as a specific instance of an APM. Empirically, Guided IG creates saliency maps better aligned with the model's prediction and the input image that is being explained. We show through qualitative and quantitative experiments that Guided IG outperforms other, related methods in nearly every experiment.

Key facts. Cited by 102 papers in the map (data through 7 Oct 2026). Builds on 24 of its 32 references that are in the map. First posted on arXiv 17 Jun 2021.

Builds on

24 of 32 references.

References · 24

  1. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,441
  2. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,690
  3. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,040
  4. 2017Axiomatic Attribution for Deep Networkscited by 3,350
  5. 2013Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Mapscited by 2,348
  6. 2013Visualizing and Understanding Convolutional Networkscited by 2,278
  7. 2017Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)cited by 1,309
  8. 2017SmoothGrad: removing noise by adding noisecited by 1,097
  9. 2014Striving for Simplicity: The All Convolutional Netcited by 983
  10. 2018RISE: Randomized Input Sampling for Explanation of Black-box Modelscited by 825
  11. 2017Network Dissection: Quantifying Interpretability of Deep Visual Representationscited by 703
  12. 2017Interpretable Explanations of Black Boxes by Meaningful Perturbationcited by 611
12 more
  1. 2018The Open Images Dataset V4: Unified image classification, object detection, and visual relationship detection at scalecited by 542
  2. 2017Interpretation of Neural Networks is Fragilecited by 443
  3. 2014Understanding Deep Image Representations by Inverting Themcited by 399
  4. 2019The many Shapley values for model explanationcited by 341
  5. 2019Understanding Deep Networks via Extremal Perturbations and Smooth Maskscited by 257
  6. 2019Explanations can be manipulated and geometry is to blamecited by 215
  7. 2016Layer-wise Relevance Propagation for Neural Networks with Local Renormalization Layerscited by 197
  8. 2016What do different evaluation metrics tell us about saliency models?cited by 184
  9. 2019XRAI: Better Attributions Through Regionscited by 116
  10. 2020Attribution in Scale and Spacecited by 61
  11. 2017Grader variability and the importance of reference standards for evaluating machine learning models for diabetic retinopathycited by 42
  12. 2020Investigating Saturation Effects in Integrated Gradientscited by 22

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 3; 2022: 8; 2023: 20; 2024: 25; 2025: 19; 2026: 27.

102 papers.

Sort:

By area:

  1. 2023Inseq: An Interpretability Toolkit for Sequence Generation Modelscited by 78
  2. 2022A Rigorous Study of Integrated Gradients Method and Extensions to Internal Neuron Attributionscited by 64
  3. 2022First is Better Than Last for Language Data Influencecited by 34
  4. 2024Gradient based Feature Attribution in Explainable AI: A Technical Reviewcited by 34
  5. 2024Manifold Integrated Gradients: Riemannian Geometry for Feature Attributioncited by 21
  6. 2023Visual Explanations via Iterated Integrated Attributionscited by 18
  7. 2023IDGI: A Framework to Eliminate Explanation Noise from Integrated Gradientscited by 16
  8. 2022On the Relationship Between Explanation and Prediction: A Causal Viewcited by 14
  9. 2024Latent Concept-based Explanation of NLP Modelscited by 13
  10. 2024IG2: Integrated Gradient on Iterative Gradient Path for Feature Attributioncited by 12
  11. 2024Visual-TCAV: Concept-based Attribution and Saliency Maps for Post-hoc Explainability in Image Classificationcited by 11
  12. 2021SoK: Machine Learning Governancecited by 9
38 more
  1. 2023Integrated Decision Gradients: Compute Your Attributions Where the Model Makes Its Decisioncited by 9
  2. 2023MAEA: Multimodal Attribution for Embodied AIcited by 9
  3. 2023Deep Integrated Explanationscited by 9
  4. 2021Attribution of Predictive Uncertainties in Classification Modelscited by 8
  5. 2022This changes to that : Combining causal and non-causal explanations to generate disease progression in capsule endoscopycited by 8
  6. 2023Gradient-based Uncertainty Attribution for Explainable Bayesian Deep Learningcited by 8
  7. 2023SMOOT: Saliency Guided Mask Optimized Online Trainingcited by 8
  8. 2023Negative Flux Aggregation to Estimate Feature Attributionscited by 7
  9. 2023MFABA: A More Faithful and Accelerated Boundary-based Attribution Method for Deep Neural Networkscited by 7
  10. 2024Post-hoc Part-prototype Networkscited by 7
  11. 2025GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMscited by 6
  12. 2023Non-Uniform Interpolation in Integrated Gradients for Low-Latency Explainable-AIcited by 5
  13. 2023LICO: Explainable Models with Language-Image Consistencycited by 5
  14. 2024Class Machine Unlearning for Complex Data via Concepts Inference and Data Poisoningcited by 5
  15. 2025Explainable AI: XAI-Guided Context-Aware Data Augmentationcited by 5
  16. 2025First is Not Really Better Than Last: Evaluating Layer Choice and Aggregation Strategies in Language Model Data Influence Estimationcited by 5
  17. 2026IGU-LoRA: Adaptive Rank Allocation via Integrated Gradients and Uncertainty-Aware Scoringcited by 5
  18. 2024Path Choice Matters for Clear Attribution in Path Methodscited by 4
  19. 2024QUCE: The Minimisation and Quantification of Path-Based Uncertainty for Generative Counterfactual Explanationscited by 4
  20. 2024What Sketch Explainability Really Means for Downstream Taskscited by 4
  21. 2024Enhancing Model Interpretability with Local Attribution over Global Explorationcited by 4
  22. 2024Reliable Evaluation of Attribution Maps in CNNs: A Perturbation-Based Approachcited by 4
  23. 2022Comparing Baseline Shapley and Integrated Gradients for Local Explanation: Some Additional Insightscited by 3
  24. 2023Towards credible visual model interpretation with path attributioncited by 3
  25. 2024Explaining Caption-Image Interactions in CLIP Models with Second-Order Attributionscited by 3
  26. 2025Using the Path of Least Resistance to Explain Deep Networkscited by 3
  27. 2025Contrastive Integrated Gradients: A Feature Attribution-Based Method for Explaining Whole Slide Image Classificationcited by 3
  28. 2023Leaping Into Memories: Space-Time Deep Feature Synthesiscited by 2
  29. 2023Insights Into the Inner Workings of Transformer Models for Protein Function Predictioncited by 2
  30. 2023COSE: A Consistency-Sensitivity Metric for Saliency on Image Classificationcited by 2
  31. 2024Challenging the Black Box: A Comprehensive Evaluation of Attribution Maps of CNN Applications in Agriculture and Forestrycited by 2
  32. 2024Forward Learning for Gradient-based Black-box Saliency Map Generationcited by 2
  33. 2024AdaptGrad: Adaptive Sampling to Reduce Noisecited by 2
  34. 2024Derivative-Free Diffusion Manifold-Constrained Gradient for Unified XAIcited by 2
  35. 2025Time series saliency maps: explaining models across multiple domainscited by 2
  36. 2025On the Complexity-Faithfulness Trade-off of Gradient-Based Explanationscited by 2
  37. 2025Mitigating Multimodal Hallucinations via Gradient-based Self-Reflectioncited by 2
  38. 2025Improving Perturbation-based Explanations by Understanding the Role of Uncertainty Calibrationcited by 2

50 most cited of 102 All 102 (JSON)