AI Paper Atlas

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

Geometrically Constrained Outlier Synthesis

Daniil Karzanov, Marcin Detyniecki

ICML 2026 · Label-efficient and robust learning › OOD and anomaly detection · approach: Uncertainty and misclassification detection · 0 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 GCOS conformal geometric synthesis instead of VOS Gaussian synthesisenergy-based OOD detector, WRN-40-2 · MVTec near-OODFPR95: 70.77 -> 23.08Improved
use GCOS instead of VOSenergy-based OOD detector · Colored MNISTFPR95: 18.50 -> 1.00Improved
use GCOS instead of VOSenergy-based OOD detector · Stanford DogsAUROC: 99.25 -> 99.55Improved
use GCOS instead of VOSenergy-based OOD detector · RetinopathyFPR95: 80.00 -> 73.00Improved
add VOS Gaussian outlier synthesis to no-regularization baselineenergy-based OOD detector · MVTec near-OODFPR95: 27.69 -> 70.77Hurt
add VOS Gaussian outlier synthesis to no-regularization baselineenergy-based OOD detector · Colored MNISTFPR95: 32.50 -> 18.50Improved
use GCOS instead of NCIS (diffusion-based)energy-based OOD detector · MVTec near-OODAUROC: 96.50 -> 95.61Hurt
use GCOS instead of NCISenergy-based OOD detector · Average over 4 datasetsAvg. AUROC: 91.97 -> 93.47Improved
replace energy-based inference with conformal hypothesis testing head (ablation)GCOS + L_uncertainty · MVTec near-OODAUROC: 94.86 -> 50.92Hurt
replace energy-based inference with conformal hypothesis testing head (ablation)GCOS + L_reg(Mahalanobis) · Colored MNISTFPR95: 1.50Improved
use energy regularization with conformal testing head (ablation)GCOS + L_reg(Energy) · MVTec near-OODAUROC: 63.67 -> 70.62Improved

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