AI Paper Atlas

Research map / Label-efficient and robust learning / Test-time adaptation

BREEDS: Benchmarks for Subpopulation Shift

Shibani Santurkar, Dimitris Tsipras, Aleksander Madry

ICLR 2021 · Label-efficient and robust learning › Test-time adaptation · approach: Characterizing out-of-distribution failures · 17 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 with ERM on source subpopulations and test on unseen target subpopulationsImageNet classifiers (AlexNet to DenseNet-121) · BREEDS taskstarget accuracyHurt
Use more accurate source modelsstandard-trained models · BREEDS tasksfraction of source accuracy preservedImproved
Retrain last linear layer on target-domain datasource-trained models · BREEDS taskstarget accuracyImproved
Adversarial training (l2 PGD)ImageNet classifiers · BREEDS tasksrelative accuracy (target/source)Improved
Erase noise augmentationImageNet classifiers · BREEDS tasksrelative accuracy (target/source)Improved
Gaussian noise augmentationImageNet classifiers · BREEDS tasksrelative accuracy (target/source)No clear effect
Stylized ImageNet trainingImageNet classifiers · BREEDS taskstarget accuracy after last-layer retrainingHurt
Adversarial training followed by last-layer retraining on targetImageNet classifiers · Living-17target accuracy after retrainingImproved
Use BREEDS-modified WordNet hierarchy instead of raw WordNet superclasses (ablation)task construction · human annotator binary tasksannotator accuracyImproved
Compare model (DenseNet-121) with humans on binary BREEDS tasksDenseNet-121 · binary BREEDS target domainaccuracyHurt
Introduce subpopulation shift for human annotators (source to target)human annotators · BREEDS binary tasksannotator accuracyNo 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