AI Paper Atlas

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

STEP: Out-of-Distribution Detection in the Presence of Limited In-Distribution Labeled Data

Zhi Zhou, Lan-Zhe Guo, Zhanzhan Cheng, Yu-Feng Li, Shiliang Pu

NeurIPS 2021 · Label-efficient and robust learning › OOD and anomaly detection · approach: Spurious-aware OOD detection · 10 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 Structure-Keep loss to unzipping (vs unzipping without it) (ablation)STEP · CIFAR-10 vs TINr / LSUNrAUROC: 79.58 -> 95.62 (TINr); 80.38 -> 99.07 (LSUNr)Improved
use projected detection-specific space (STEP) vs plain Mahalanobis distance (ablation)STEP · CIFAR-10 vs TINr / LSUNrAUROC: 90.96 -> 95.62 (TINr); 93.46 -> 99.07 (LSUNr)Improved
add KNN-based geodesic distance to Mahalanobis (ablation)MAH · CIFAR-10 vs TINr / LSUNrAUROC: 90.96 -> 91.26 (TINr); 93.46 -> 97.35 (LSUNr)Mixed
STEP vs UOOD† (uses OOD validation set)STEP · CIFAR-100 vs TINr, LSUNrAUROC: 95.31 -> 93.51 (TINr); 98.96 -> 98.20 (LSUNr)Hurt
STEP vs UOOD† on CIFAR-10 benchmarksSTEP · CIFAR-10 vs TINrAUROC: 92.63 -> 95.61Improved
train with known OOD set then test on unknown OOD setUOOD† and MAH† · CIFAR-10 known vs unknown OODAUROCHurt
train with known OOD set then test on unknown OOD setSTEP · CIFAR-10 known vs unknown OODAUROCNo clear effect
vary KNN neighbors K from 2 to 18 (ablation)STEP · various OOD setsAUROCNo 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