AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Learning to Estimate Shapley Values with Vision Transformers

Ian Connick Covert, Chanwoo Kim, Su-In Lee

ICLR 2023 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 12 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
Train ViT with random attention masking / fine-tune with masking (vs no fine-tuning) (ablation)ViT classifier handling held-out patches · ImageNette, MURAKL divergence to full-image prediction under patch removalImproved
Use learned ViT explainer (ViT Shapley) instead of baselinesViT explanations, target class · ImageNette and MURA insertion/deletion/faithfulnessinsertion, deletion, faithfulnessImproved
Explain non-target classes with ViT Shapley vs baselinesViT explanations, non-target classes · ImageNetteinsertion and faithfulnessImproved
Add extra attention block and full fine-tuning to explainer (config D vs config A) (ablation)ViT Shapley explainer architecture · ImageNettevalidation loss: 4.309 vs 4.332Improved
Train explainer from scratch instead of fine-tuning ViT (ablation)ViT Shapley explainer · ImageNettevalidation loss: 4.331 vs 4.309 for best fine-tuned configHurt
Use attention-based explanations (attention last, rollout) for ViTViT explanations · ImageNette non-target classesinsertion/deletion/faithfulnessNo clear effect
Gradient-based methods (GradCAM, SmoothGrad, VarGrad) vs random baselineViT explanations, target class · ImageNette insertioninsertionNo clear effect

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