AI Paper Atlas

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

Contrastive Adapters for Foundation Model Group Robustness

Michael Zhang, Christopher Re

NeurIPS 2022 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Robust fine-tuning of zero-shot models · 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
Replace zero-shot inference with contrastive adapter trainingCLIP ResNet-50 · Waterbirdsworst-group accuracy: 36.6 -> 83.7Improved
Replace zero-shot inference with contrastive adapter trainingCLIP ResNet-50 · CelebAworst-group accuracy: 74.0 -> 90.0Improved
Replace zero-shot inference with contrastive adapter trainingCLIP ResNet-50 · BREEDS Living-17worst-group accuracy: 6.0 -> 62.0Improved
Replace zero-shot inference with contrastive adapter trainingCLIP ResNet-50 · CIFAR-10.02worst-group accuracy: 39.1 -> 60.7Improved
Train an ERM adapter instead of using zero-shot inferenceCLIP ResNet-50 · CelebAworst-group accuracy: 74.0 -> 36.1Hurt
Train an ERM adapter instead of using zero-shot inferenceCLIP ResNet-50 · Waterbirdsworst-group accuracy: 36.6 -> 60.8Improved
Train an ERM adapter instead of using zero-shot inferenceCLIP ResNet-50 · BREEDS Living-17worst-group accuracy: 6.0 -> 70.7Improved
Train an ERM adapter instead of using zero-shot inferenceCLIP ResNet-50 · CIFAR-10.02worst-group accuracy: 39.1 -> 68.8Improved
Use contrastive adapter training rather than ERM adapter trainingCLIP ViT-L/14 · CelebAworst-group accuracy: 36.7 -> 84.6Improved
Train a linear probe instead of using zero-shot inferenceCLIP ResNet-50 · CelebAworst-group accuracy: 74.0 -> 11.9Hurt
Add a batch normalization layer between adapter input and output layers (ablation)Adapter · Not specifiedgroup robustnessImproved
Add a residual connection to the adapter (ablation)Adapter · Not specifiedgroup robustnessNo 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