AI Paper Atlas

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

Self-Supervised Learning for Generalizable Out-of-Distribution Detection

Sina Mohseni, Mandar Pitale, JBS Yadawa, Zhangyang Wang

AAAI 2020 · Label-efficient and robust learning › OOD and anomaly detection · approach: Spurious-aware OOD detection · 160 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
add self-supervised reject-class head trained on unlabeled OOD data instead of OE calibrationWRN classifier · CIFAR-100 (avg 5 OOD sets)AUROC: +6.89%Improved
same, vs OEWRN classifier · CIFAR-10 (avg 5 OOD sets)AUROC: +1.80%Improved
same, vs OEclassifier · SVHN / MNISTAUROC: +0.21% / +1.29%Improved
evaluate on mixture of OOD distributionsODIN / Mahalanobis (tuned on target OOD) · CIFAR-100AUROC: -14.07% (ODIN), -25.84% (Mahalanobis)Hurt
use ImageNet-22k instead of Tiny Images as OOD training set (ablation)Ours · CIFAR-10AUROC: 97.37% -> 90.40%Hurt
blend ImageNet OOD samples with in-distribution images (alpha 0.1) (ablation)Ours with ImageNet OOD · CIFAR-10AUROC: 90.40% -> 91.72%Improved
vary number of reject classes k (ablation)Ours · CIFAR-10AUROC: std 0.31%No clear effect
vary number of reject classes k (ablation)Ours · CIFAR-100AUROC: std 5.40%Mixed
use alternative OOD scores (sum, entropy) instead of max softmax of reject classes (ablation)Ours · CIFAR-10/100 mixed OODAUROC: 97.77 -> 97.88 (entropy)No clear effect
use the OOD detector for selective classificationWRN classifier · CIFAR-100normal classification error: -11.81%Improved
Ours vs Deep SVDDone-class classification · CIFAR-10AUROC: 64.81% -> 77.75%Improved

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