AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

The Cells Out of Sample (COOS) dataset and benchmarks for measuring out-of-sample generalization of image classifiers

Alex Lu, Amy Lu, Wiebke Schormann, Marzyeh Ghassemi, David Andrews, Alan Moses

NeurIPS 2019 · Label-efficient and robust learning › Recognition and long-tail · approach: Group-robust classification · 14 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
Move from in-distribution Test1 to months-later Test3DeepLoc end-to-end CNN · COOS-7class-balanced error (%): 1.2 -> 7.4Hurt
Move from Test1 to different-microscope Test4DeepLoc end-to-end CNN · COOS-7class-balanced error (%): 1.2 -> 5.4Hurt
Move from Test1 to Test3Texture L1 LR · COOS-7class-balanced error (%): 6.8 -> 12.0Hurt
Move from Test1 to Test4Texture kNN · COOS-7class-balanced error (%): 11.8 -> 25.6Hurt
Use kNN on last-layer DeepLoc features instead of the CNN classifierDeepLoc · COOS-7 Test3class-balanced error (%): 7.4 -> 6.9No clear effect
Replace kNN with L1 logistic regression on Texture featuresTexture · COOS-7 Test3class-balanced error (%): 17.6 -> 12.0Improved
Replace kNN with L1 logistic regression on Texture featuresTexture · COOS-7 Test4class-balanced error (%): 25.6 -> 12.1Improved
Replace kNN with L1 logistic regression on DeepLoc featuresDeepLoc (FC2) · COOS-7 Test3class-balanced error (%): 6.9 -> 7.7No clear effect
Use random forest instead of kNN on VGG16 Conv4_2 featuresVGG16 · COOS-7 Test1class-balanced error (%): 7.8 -> 7.4Improved
Use deeper PCI layer (Conv5 vs Conv3) with L1 LRPCI · COOS-7 Test3class-balanced error (%): 9.2 -> 11.4Hurt
Use deeper PCI layer (Conv5 vs Conv3) with L1 LRPCI · COOS-7 Test4class-balanced error (%): 7.4 -> 5.7Improved
Use VGG16 Conv4_2 instead of Conv3_3 features with L1 LRVGG16 · COOS-7 Test3class-balanced error (%): 11.3 -> 8.0Improved

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