mlsanityData to 7 Oct 2026

DAVE: Distribution-aware Attribution via ViT Gradient Decomposition

arXiv 2602.06613 ↗First posted 6 Feb 2026cs.CV cs.AI cs.HCcited by 3

Not on the map. Why?

Abstract (arXiv)

Vision Transformers (ViTs) have become a dominant architecture in computer vision, yet producing stable and high-resolution attribution maps for these models remains challenging. Architectural components such as patch embeddings and attention routing often introduce structured artifacts in pixel-level explanations, causing many existing methods to rely on coarse patch-level attributions. We introduce DAVE textit(underlineDistribution-aware underlineAttribution via underlineViT Gradient DunderlineEcomposition), a mathematically grounded attribution method for ViTs based on a structured decomposition of the input gradient. By exploiting architectural properties of ViTs, DAVE isolates locally equivariant and stable components of the effective input--output mapping. It separates these from architecture-induced artifacts and other sources of instability.

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 30 of its 34 references that are in the map. First posted on arXiv 6 Feb 2026.

Builds on

30 of 34 references.

References · 30

  1. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  2. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,441
  3. 2021Emerging Properties in Self-Supervised Vision Transformerscited by 6,893
  4. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,690
  5. 2020Training data-efficient image transformers & distillation through attentioncited by 4,560
  6. 2017Axiomatic Attribution for Deep Networkscited by 3,350
  7. 2013Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Mapscited by 2,348
  8. 2017Learning Important Features Through Propagating Activation Differencescited by 1,454
  9. 2017SmoothGrad: removing noise by adding noisecited by 1,097
  10. 2021Transformers in Vision: A Surveycited by 1,009
  11. 2020Quantifying Attention Flow in Transformerscited by 759
  12. 2018This Looks Like That: Deep Learning for Interpretable Image Recognitioncited by 731
18 more
  1. 2020Transformer Interpretability Beyond Attention Visualizationcited by 620
  2. 2019Score-CAM: Score-Weighted Visual Explanations for Convolutional Neural Networkscited by 575
  3. 2022DeiT III: Revenge of the ViTcited by 417
  4. 2016Top-down Neural Attention by Excitation Backpropcited by 318
  5. 2019Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNetcited by 260
  6. 2019Full-Gradient Representation for Neural Network Visualizationcited by 189
  7. 2024AttnLRP: Attention-Aware Layer-Wise Relevance Propagation for Transformerscited by 116
  8. 2022B-cos Networks: Alignment is All We Need for Interpretabilitycited by 88
  9. 2020There and Back Again: Revisiting Backpropagation Saliency Methodscited by 60
  10. 2021Convolutional Dynamic Alignment Networks for Interpretable Classificationscited by 54
  11. 2024LeGrad: An Explainability Method for Vision Transformers via Feature Formation Sensitivitycited by 32
  12. 2023B-cos Alignment for Inherently Interpretable CNNs and Vision Transformerscited by 31
  13. 2023Towards Evaluating Explanations of Vision Transformers for Medical Imagingcited by 22
  14. 2023Visual Explanations via Iterated Integrated Attributionscited by 18
  15. 2024B-cosification: Transforming Deep Neural Networks to be Inherently Interpretablecited by 12
  16. 2025Artifacts and Attention Sinks: Structured Approximations for Efficient Vision Transformerscited by 9
  17. 2025How to Probe: Simple Yet Effective Techniques for Improving Post-hoc Explanationscited by 8
  18. 2024OMENN: One Matrix to Explain Neural Networkscited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026Conceptualizing Embeddings: Sparse Disentanglement for Vision-Language Modelscited by 0
  2. 2026What Pixels Are Enough? SEAMS: Sufficiency Saliency via MSE-Preservation Soft-Maskscited by 0
  3. 2026ResLRP: The Role of Residual Cancellation in Attribution Instability in Vision Transformerscited by 0