AI Paper Atlas

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

Meta OOD Learning For Continuously Adaptive OOD Detection

Xinheng Wu, Jie Lu, Zhen Fang, Guangquan Zhang

ICCV 2023 · Label-efficient and robust learning › OOD and anomaly detection · approach: Provable and feature-based OOD detection · 8 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 uncertainty regularization L_ood with virtual OOD samples (ablation)MOL · R-MNISTAUROC (R-NOTMNIST): 94.1 -> 96.5Improved
add distribution discrepancy constraint L_qry in outer loop (ablation)MOL · R-MNISTID accuracy: 32.2 -> 35.6Improved
start L_ood earlier in training (E=1/3 instead of 2/3) (ablation)MOL · R-MNISTID accuracy: 35.6 -> 28.7Hurt
start L_ood later in training (E=4/5 instead of 2/3) (ablation)MOL · R-MNISTAUROC (R-Cifar10bw): 98.9 -> 95.2Hurt
use random sampled test trajectories instead of fixed (ablation)MOL · R-MNISTAUROC (R-NOTMNIST): 97.9 -> 97.7 ± 0.4No clear effect
fine-tune classifier at test time (Simple Adaptive) for static OOD baselinesstatic OOD baselines · R-MNISTID accuracy / FPR95: around 2.7% ID accuracy gain; FPR95 reduced 5.3% to 19.3%Improved
train baselines with domain adaptation (DAN) using all shifting training samplesstatic OOD baselines · R-MNISTAUROC: AUROC dropped 17.9% to 26.3%Hurt
MOL versus best static baselines with adaptationMOL · Cifar10Caverage AUROC: 66.4 -> 70.6Improved
MOL versus Gram on R-MNIST FPR95MOL · R-MNISTaverage FPR95: 7.6 vs 11.5Hurt

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