AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Transformer Interpretability Beyond Attention Visualization

Hila Chefer, Shir Gur, Lior Wolf

CVPR 2021 · Architectures and efficiency › Attention and transformers · approach: Attention head and block redesign · 838 citations (OpenAlex)

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use class-specific relevance and gradient propagation integrated across attention blocksViT-base explanations · ImageNet-SegmentationmIoU: 57.94 -> 61.95Improved
Use class-specific relevance and gradient propagation integrated across attention blocksViT-base explanations · ImageNet val, negative perturbation, predicted classAUC: 53.1 -> 54.16Improved
Use class-specific relevance and gradient propagation integrated across attention blocksViT-base explanations · ImageNet val, positive perturbation, predicted classAUC: 19.64 -> 17.03Improved
Use class-specific relevance and gradient propagation integrated across attention blocksViT-base explanations · ImageNet val, negative perturbation, target classAUC: 50.49 -> 55.04Improved
Use class-specific relevance and gradient propagation integrated across attention blocksViT-base explanations · ImageNet val, positive perturbation, target classAUC: 19.64 -> 16.04Improved
Replace attention-map gradients with attention maps (ablation)Full method · ImageNet-Segmentationsegmentation performanceHurt
Add rollout across attention blocks rather than use only the output-nearest block (ablation)Full method · ImageNet-Segmentation and ImageNet predicted-class perturbationperformanceImproved
Use the output-nearest attention block rather than the input-nearest block (ablation)Single-block relevance-gradient explanation · ImageNet visualizationsinformativenessImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy