mlsanityData to 14 Oct 2026

VISION DIFFMASK: Faithful Interpretation of Vision Transformers with Differentiable Patch Masking

arXiv 2304.06391 ↗First posted 13 Apr 2023cs.CV cs.AI cs.LGcited by 4

On the map: Mechanistic interpretability → Interpreting vision and multimodal models

Abstract (arXiv)

The lack of interpretability of the Vision Transformer may hinder its use in critical real-world applications despite its effectiveness. To overcome this issue, we propose a post-hoc interpretability method called VISION DIFFMASK, which uses the activations of the model's hidden layers to predict the relevant parts of the input that contribute to its final predictions. Our approach uses a gating mechanism to identify the minimal subset of the original input that preserves the predicted distribution over classes. We demonstrate the faithfulness of our method, by introducing a faithfulness task, and comparing it to other state-of-the-art attribution methods on CIFAR-10 and ImageNet-1K, achieving compelling results. To aid reproducibility and further extension of our work, we open source our implementation: https://github.com/AngelosNal/Vision-DiffMask

Key facts. Cited by 4 papers in the map (data through 14 Oct 2026). Placed in Interpreting vision and multimodal models, one of 13 directions in Mechanistic interpretability. Builds on 23 of its 28 references that are in the map. First posted on arXiv 13 Apr 2023.

Builds on

23 of 28 references.

References · 23

  1. 2017Attention Is All You Needcited by 60,361
  2. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  3. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
  4. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,070
  5. 2017Axiomatic Attribution for Deep Networkscited by 3,384
  6. 2021Convex optimizationcited by 2,515
  7. 2013Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Mapscited by 2,371
  8. 2017Learning Important Features Through Propagating Activation Differencescited by 1,458
  9. 2019Attention is not Explanationcited by 929
  10. 2018RISE: Randomized Input Sampling for Explanation of Black-box Modelscited by 838
  11. 2020Quantifying Attention Flow in Transformerscited by 775
  12. 2019Attention is not not Explanationcited by 636
11 more
  1. 2020Transformer Interpretability Beyond Attention Visualizationcited by 627
  2. 2017Interpretable Explanations of Black Boxes by Meaningful Perturbationcited by 612
  3. 2020Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?cited by 599
  4. 2017Learning Sparse Neural Networks through L0 Regularizationcited by 583
  5. 2019Lookahead Optimizer: k steps forward, 1 step backcited by 336
  6. 2016Top-down Neural Attention by Excitation Backpropcited by 321
  7. 2017Real Time Image Saliency for Black Box Classifierscited by 246
  8. 2016Understanding Neural Networks through Representation Erasurecited by 224
  9. 2016Layer-wise Relevance Propagation for Neural Networks with Local Renormalization Layerscited by 199
  10. 2019Interpretable Neural Predictions with Differentiable Binary Variablescited by 126
  11. 2020How do Decisions Emerge across Layers in Neural Models? Interpretation with Differentiable Maskingcited by 63

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 2; 2024: 0; 2025: 0; 2026: 2.

4 papers.

  1. 2023Explainability of Vision Transformers: A Comprehensive Review and New Perspectivescited by 12
  2. 2023Adventures of Trustworthy Vision-Language Models: A Surveycited by 10
  3. 2026What Pixels Are Enough? SEAMS: Sufficiency Saliency via MSE-Preservation Soft-Maskscited by 0
  4. 2026Grad-CAM for Vision Transformers: A Systematic Taxonomy and Audit of Methodological Ambiguity in Explainable AIcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2023Interpretability-Aware Vision Transformer
  2. 2026Can Cross-Layer Transcoders Replace Vision Transformer Activations? An Interpretable Perspective on Vision
  3. 2026Sparse but not Simpler: A Multi-Level Interpretability Analysis of Vision Transformers
  4. 2025[Re] Improving Interpretation Faithfulness for Vision Transformers
  5. 2024An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels
  6. 2025Interpreting vision transformers via residual replacement model
  7. 2024ADFQ-ViT: Activation-Distribution-Friendly Post-Training Quantization for Vision Transformers
  8. 2026Decoding the Functional Roles of Register and High-Norm Patch Tokens in Vision Transformers