mlsanityData to 14 Oct 2026

Holistically Explainable Vision Transformers

arXiv 2301.08669 ↗First posted 20 Jan 2023cs.CV stat.MLcited by 8

Not on the map. Why?

Abstract (arXiv)

Transformers increasingly dominate the machine learning landscape across many tasks and domains, which increases the importance for understanding their outputs. While their attention modules provide partial insight into their inner workings, the attention scores have been shown to be insufficient for explaining the models as a whole. To address this, we propose B-cos transformers, which inherently provide holistic explanations for their decisions. Specifically, we formulate each model component - such as the multi-layer perceptrons, attention layers, and the tokenisation module - to be dynamic linear, which allows us to faithfully summarise the entire transformer via a single linear transform. We apply our proposed design to Vision Transformers (ViTs) and show that the resulting models, dubbed Bcos-ViTs, are highly interpretable and perform competitively to baseline ViTs on ImageNet. Code will be made available soon.

Key facts. Cited by 8 papers in the map (data through 14 Oct 2026). Builds on 33 of its 42 references that are in the map. First posted on arXiv 20 Jan 2023.

Builds on

33 of 42 references.

References · 33

  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. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
  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. 2016Layer Normalizationcited by 5,669
  7. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,070
  8. 2016Gaussian Error Linear Units (GELUs)cited by 3,502
  9. 2017Axiomatic Attribution for Deep Networkscited by 3,384
  10. 2015Learning Deep Features for Discriminative Localizationcited by 2,436
  11. 2013Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Mapscited by 2,371
  12. 2019RandAugment: Practical automated data augmentation with a reduced search spacecited by 2,269
21 more
  1. 2017Learning Important Features Through Propagating Activation Differencescited by 1,458
  2. 2017Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)cited by 1,319
  3. 2018Sanity Checks for Saliency Mapscited by 1,156
  4. 2014Striving for Simplicity: The All Convolutional Netcited by 990
  5. 2019Attention is not Explanationcited by 929
  6. 2019Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Prunedcited by 856
  7. 2018RISE: Randomized Input Sampling for Explanation of Black-box Modelscited by 838
  8. 2020Quantifying Attention Flow in Transformerscited by 775
  9. 2017Network Dissection: Quantifying Interpretability of Deep Visual Representationscited by 708
  10. 2020Transformer Interpretability Beyond Attention Visualizationcited by 627
  11. 2021How to train your ViT? Data, Augmentation, and Regularization in Vision Transformerscited by 569
  12. 2018Towards Robust Interpretability with Self-Explaining Neural Networkscited by 505
  13. 2021LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inferencecited by 475
  14. 2021Early Convolutions Help Transformers See Bettercited by 396
  15. 2019Is Attention Interpretable?cited by 389
  16. 2019Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNetcited by 260
  17. 2019Full-Gradient Representation for Neural Network Visualizationcited by 191
  18. 2013Maxout Networkscited by 187
  19. 2020The elephant in the interpretability room: Why use attention as explanation when we have saliency methods?cited by 122
  20. 2021Convolutional Dynamic Alignment Networks for Interpretable Classificationscited by 54
  21. 2022Grad-SAM: Explaining Transformers via Gradient Self-Attention Mapscited by 42

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 3; 2024: 2; 2025: 2; 2026: 1.

8 papers.

  1. 2023Artificial General Intelligence for Medical Imaging Analysiscited by 32
  2. 2023Transformers in Healthcare: A Surveycited by 17
  3. 2023Explainability of Vision Transformers: A Comprehensive Review and New Perspectivescited by 12
  4. 2024Benchmarking the Attribution Quality of Vision Modelscited by 8
  5. 2024T-TAME: Trainable Attention Mechanism for Explaining Convolutional Networks and Vision Transformerscited by 4
  6. 2025QPM: Discrete Optimization for Globally Interpretable Image Classificationcited by 2
  7. 2025Detection Transformers Under the Knife: A Neuroscience-Inspired Approach to Ablationscited by 0
  8. 2026qZACH-ViT: Quantization-Aware Intrinsic Explanations with Recursive Attribution-Stabilized Optimizationcited by 0