AI Paper Atlas

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

How Out-of-Distribution Detection Learning Theory Enhances Transformer: Learnability and Reliability

Yijin Zhou, Yutang Ge, Wenyuan XIE, Linqian Zeng, Xiaowen Dong, Yu Guang Wang

ICML 2026 · Label-efficient and robust learning › OOD and anomaly detection · approach: Provable and feature-based OOD 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
add synthetic rounded OOD data generation during fine-tuning (ablation)transformer on Gaussian mixture data · Gaussian mixture toy dataID/OOD accuracyImproved
add ID-OOD binary classification loss L2 without synthetic outliers (ablation)cross-entropy-trained transformer · Gaussian mixture toy dataOOD detectionNo clear effect
increase weight gamma of the ID-OOD loss (ablation)GROD loss · Gaussian mixture toy dataID accuracy vs OOD recognition: gamma 0.0, 0.5, 0.9Mixed
replace cross-entropy-only training with GROD (L2 plus synthetic soft OOD) on DINODINO transformer · CIFAR-10 average FPR@95FPR@95 (%): 5.41 -> 3.01 (vs VIM)Improved
apply GROD instead of best baselineDINO transformer · CIFAR-100 average FPR@95FPR@95 (%): 24.28 -> 18.85Improved
apply GROD instead of best baselineDINO transformer · ImageNet-200 averageFPR@95 (%): 33.05 -> 30.90Improved
apply GROD to LoRA fine-tuned Llama-3.1-8BLlama-3.1-8B · IMDB vs YelpAUROC (%): 57.66 -> 67.00Improved
apply GROD to LoRA fine-tuned Llama-3.1-8BLlama-3.1-8B · CLINC150 unknown intentsAUROC (%): 93.69 -> 93.64No clear effect
fine-tune with GROD versus NPOS and CIDER on textLlama-3.1-8B · IMDB vs YelpFPR@95 (%): 99.99 -> 83.28Improved

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