AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Generic Attention-Model Explainability for Interpreting Bi-Modal and Encoder-Decoder Transformers

Hila Chefer, Shir Gur, Lior Wolf

ICCV 2021 · Architectures and efficiency › Attention and transformers · approach: Theory of self-attention dynamics · 361 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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
add self-attention normalization to bi-modal rule (ablation)explainability method on DETR · COCO val weak segmentationAP: 11.7 -> 13.1Improved
add self-attention normalization to bi-modal rule (ablation)explainability method on DETR · COCO val weak segmentationAP_large: 20.9 -> 24.6Improved
add self-attention normalization (ablation)explainability method on DETR · COCO val weak segmentationAR_medium: 23.9 -> 23.9No clear effect
add aggregation of previous relevancy matrices (ablation)explainability method on DETR · COCO val weak segmentationAP: 0.1 -> 13.1Improved
add self-attention terms to bi-modal rule (ablation)explainability method on DETR · COCO val weak segmentationAP: 11.5 -> 13.1Improved
add normalization (ablation)explainability method on LXMERT · VQA perturbationnegative AUC, image tokens: 62.49 -> 63.24Improved
add aggregation (ablation)explainability method on LXMERT · VQA perturbationnegative AUC, image tokens: 60.41 -> 63.24Improved
add aggregation (ablation)explainability method on LXMERT · VQA perturbationpositive AUC, image tokens (lower better): 60.40 -> 51.10Improved
add self-attention terms to bi-modal rule (ablation)explainability method on LXMERT · VQA perturbationnegative AUC, image tokens: 62.18 -> 63.24Improved
add LRP to head averagingexplainability method on LXMERT · VQA perturbationnegative AUC, image tokens: 63.24 -> 62.41Hurt
add LRP to head averagingexplainability method on LXMERT · VQA perturbationnegative AUC, text tokens: 48.70 -> 48.25Hurt
use proposed method instead of Transformer attributionDETR · COCO val weak segmentationAP: 7.2 -> 13.1Improved

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