AI Paper Atlas

Research map / Label-efficient and robust learning / Noisy labels

Identifying Mislabeled Data using the Area Under the Margin Ranking

Geoff Pleiss, Tianyi Zhang, Ethan Elenberg, Kilian Q. Weinberger

NeurIPS 2020 · Label-efficient and robust learning › Noisy labels · approach: Loss correction and noise robustness theory · 45 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
Use AUM to identify and remove mislabeled samplesResNet-32 · CIFAR10, uniform noisetest-error: 9.8 ± 0.0 vs. 25.0 ± 0.3; 12.5 ± 0.1 vs. 43.3 ± 0.4; 17.9 ± 0.0 vs. 63.3 ± 0.4; 45.6 ± 1.6 vs. 83.4 ± 0.2Improved
Use AUM to identify and remove mislabeled samplesResNet-32 · CIFAR100, uniform noisetest-error: 34.5 ± 0.2 vs. 50.4 ± 0.2; 38.7 ± 0.1 vs. 62.5 ± 0.2; 47.0 ± 0.5 vs. 76.2 ± 0.4; 68.3 ± 0.7 vs. 91.8 ± 0.3Improved
Use AUM to identify and remove mislabeled samplesResNet-32 · Tiny ImageNet, uniform noisetest-error: 51.1 ± 0.2 vs. 58.6 ± 0.1; 55.3 ± 0.1 vs. 66.3 ± 0.2; 62.7 ± 0.3 vs. 77.6 ± 0.1; 79.2 ± 0.2 vs. 92.4 ± 0.1Improved
Remove AUM-identified training samplesResNet-50 · WebVision50test-error: 21.4 -> 19.8Improved
Remove AUM-identified training samplesResNet-50 · Clothing100Ktest-error: 35.8 -> 33.5Improved
Remove AUM-identified training samplesResNet-32 · CIFAR100test-error: 33.0 -> 31.8Improved
Remove AUM-identified training samplesResNet-50 · ImageNettest-error: 24.2 -> 24.4Hurt
Rank samples by AUM rather than by a single-epoch margin (ablation)AUM · CIFAR10, 40% noiseclean/mislabeled sample separationImproved
Discard samples randomly instead of by AUM ranking (ablation)ResNet-32 · CIFAR100, 0% and 40% uniform noisetest-errorHurt
Choose the removal threshold using the 99th percentile of threshold-sample AUM (ablation)AUM · CIFAR100, 0% and 40% uniform noisetest-errorImproved
Use AUM without data augmentation (ablation)ResNet-32 · CIFAR10, 40% noisetest-error: 51% -> 20%Improved
Use AUM with data augmentation (ablation)ResNet-32 · CIFAR10, 40% noisetest-error: 43% -> 12%Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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