AI Paper Atlas

Research map / Label-efficient and robust learning / OOD and anomaly detection

Breaking Down Out-of-Distribution Detection: Many Methods Based on OOD Training Data Estimate a Combination of the Same Core Quantities

Julian Bitterwolf, Alexander Meinke, Maximilian Augustin, Matthias Hein

ICML 2022 · Label-efficient and robust learning › OOD and anomaly detection · approach: Provable and feature-based OOD detection · 5 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
combine a shared in/out discriminator with the classifier using s2Shared Combi · CIFAR-10; test OOD datasetsMean FPR@95%TPR: 29.34 -> 16.00Improved
combine a shared in/out discriminator with the classifier using s2Shared Combi · CIFAR-100; test OOD datasetsMean FPR@95%TPR: 67.23 -> 33.01Improved
use a shared representation for the classifier and discriminator instead of separate representations (ablation)classifier plus binary discriminator · CIFAR-10; test OOD datasetsMean FPR@95%TPR: 21.40 -> 16.06Improved
use a shared representation for the classifier and discriminator instead of separate representations (ablation)classifier plus binary discriminator · CIFAR-100; test OOD datasetsMean FPR@95%TPR: 41.94 -> 33.06Improved
use a shared-trained discriminator with the separately trained plain classifier (ablation)Plain × Sha Disc · CIFAR-10; test OOD datasetsMean FPR@95%TPR: 21.40 -> 15.96Improved
use a shared-trained discriminator with the separately trained plain classifier (ablation)Plain × Sha Disc · CIFAR-100; test OOD datasetsMean FPR@95%TPR: 41.94 -> 33.38Improved
use s2 rather than s1 to score the background-class model (ablation)BGC · CIFAR-10; test OOD datasetsMean FPR@95%TPR: 18.83 -> 16.52Improved
use s2 rather than s3 to score the shared combination (ablation)Shared Combi · CIFAR-10; test OOD datasetsMean FPR@95%TPR: 16.06 -> 16.00No clear effect
use s2 rather than s3 to score the shared combination (ablation)Shared Combi · CIFAR-100; test OOD datasetsMean FPR@95%TPR: 33.06 -> 33.01No clear effect
train the binary discriminator against uniform noise instead of a diverse surrogate OOD setBinDisc Uniform · CIFAR-10 vs. unseen OOD datasetsMean AUC: 57.97 -> 45.90Hurt
use shared training with a discriminator rather than plain classifier training (ablation)Shared Classi · CIFAR-10; test OOD datasetsMean FPR@95%TPR: 53.01 -> 29.34Improved

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