AI Paper Atlas

Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning

Patching open-vocabulary models by interpolating weights

Gabriel Ilharco, Mitchell Wortsman, Samir Yitzhak Gadre, Shuran Song, Hannaneh Hajishirzi, Simon Kornblith, Ali Farhadi, Ludwig Schmidt

NeurIPS 2022 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Transductive and semantic-dictionary zero-shot · 26 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
interpolate weights of fine-tuned and zero-shot model (PAINT) instead of using the zero-shot modelCLIP ViT-L/14 · nine patching tasksaccuracy: +0.8 to +19.4 (Table 1 held-out classes); abstract: 15 to 60 points on tasksImproved
patch on task A, measure disjoint-class task BPAINT ViT-L/14 · EuroSAT/RESISC45 pairaccuracy: 71.0 -> 69.7Hurt
patch on task A, measure related task BPAINT ViT-L/14 · seven other task pairsaccuracy: +1.1 to +19.2Improved
scale up model size (ablation)PAINT · patching effectivenessaccuracy distance to optimalImproved
use joint patching versus unpatched (ablation)ViT-L/14 · nine patching tasks + ImageNetcombined accuracy: +15.8 pointsImproved
joint patching versus specialized models (ablation)ViT-L/14 · combined accuracycombined accuracy: 0.5 points worseNo clear effect
add weight interpolation to sequential fine-tuning (ablation)ViT-B/32 sequential patching · combined accuracycombined accuracy: -4.6 -> +11 points vs unpatchedImproved
parallel patching instead of joint (ablation)PAINT multi-task · nine taskscombined accuracyHurt
black-box optimization of mixing coefficients versus shared coefficient (ablation)parallel patching · nine tasksaccuracy: 0.3 to 0.4 pointsNo clear effect
use ResNets instead of ViTs (ablation)PAINT · patching effectivenessaccuracy trade-offHurt
patch on synthetic typographic attacksViT-L/14 · real-world typographic attacksaccuracy: +41 pointsImproved
patch on counting subset of classesViT-L/14 · CLEVR unseen countsaccuracy: 59% -> over 99%Improved

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