Research map / Label-efficient and robust learning / OOD and anomaly detection
Deep Anomaly Detection with Outlier Exposure
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| Fine-tune MSP classifiers with OE using auxiliary outlier data | MSP image classifiers · SVHN, CIFAR-10, CIFAR-100, Tiny ImageNet, Places365 | AUROC: 98.0 -> 100.0; 89.3 -> 97.8; 73.1 -> 87.9; 64.9 -> 92.2; 66.5 -> 90.6 | Helped |
| Fine-tune MSP text classifiers with OE | MSP text classifiers · 20 Newsgroups, TREC, SST | AUROC: 82.7 -> 97.7; 82.1 -> 99.3; 61.6 -> 89.3 | Helped |
| Apply OE to the Confidence Branch detector | Confidence Branch · CIFAR-10, CIFAR-100, Tiny ImageNet | AUROC: 86.9 -> 93.7; 81.2 -> 85.5; 63.4 -> 90.6 | Helped |
| Add OE with real, diverse data after GAN-based outlier training | MSP + GAN · CIFAR-10 and CIFAR-100 | AUROC: 89.6 -> 97.2; 69.3 -> 77.9 | Helped |
| Fine-tune PixelCNN++ with OE margin loss | PixelCNN++ · CIFAR-10 OOD detection, mean across test distributions | AUROC: 65.8 -> 83.7 | Helped |
| Fine-tune Penn Treebank language models with OE | QRNN language models · PTB Characters and PTB Words | AUROC: 77.5 -> 86.3; 81.2 -> 99.2 | Helped |
| Use GAN-generated synthetic outliers instead of OE data (ablation) | MSP OOD detector · CIFAR-10 and CIFAR-100 | AUROC: 88.1 -> 89.6; 67.2 -> 69.3 | Mixed |
| Train with OE from scratch rather than fine-tuning with OE (ablation) | CIFAR-10 Wide ResNet · CIFAR-10 OOD detection | FPR95: 9.50% -> 6.15% | Helped |
| Use the uniform-posterior entropy score instead of MSP scoring (ablation) | OE-fine-tuned classifiers · Tiny ImageNet OOD detection | AUROC: 92.18 -> 95.45 | Helped |
| Increase the number of CIFAR-100 OE outlier classes (ablation) | CIFAR-10 classifier · CIFAR-10 OOD detection | AUPR: 78.5% -> 85.1% -> 85.3% | Helped |
| Use only 50,000 examples from 80 Million Tiny Images (ablation) | OE auxiliary dataset · Image OOD detection | detection performance | No clear effect |
| Replace WikiText-2 with Project Gutenberg as OE data (ablation) | SST OOD detector · SST | AUROC: 89.3% -> 90.1% | Helped |
Closest papers by meaning
- ALTBI: Constructing Improved Outlier Detection Models via Optimization of Inlier-Memorization Effect (AAAI 2025 · similarity 0.90)
- RODEO: Robust Outlier Detection via Exposing Adaptive Out-of-Distribution Samples (ICML 2024 · similarity 0.90)
- ODIM: Outlier Detection via Likelihood of Under-Fitted Generative Models (ICML 2024 · similarity 0.89)
- DOS: Diverse Outlier Sampling for Out-of-Distribution Detection (ICLR 2024 · similarity 0.87)
- Envisioning Outlier Exposure by Large Language Models for Out-of-Distribution Detection (ICML 2024 · similarity 0.87)
- Out-Of-Distribution Detection with Diversification (Provably) (NeurIPS 2024 · similarity 0.87)
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.