Research map / Label-efficient and robust learning / Recognition and long-tail
Training Confidence-calibrated Classifiers for Detecting Out-of-Distribution Samples
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 |
|---|---|---|---|
| Add the confidence loss using CIFAR-10 as training OOD data | VGGNet trained on SVHN · SVHN vs CIFAR-10, TinyImageNet, LSUN, and Gaussian | TNR at TPR 95%: 47.4 / 49.0 / 46.3 / 56.1 -> 99.9 / 100.0 / 100.0 / 100.0 | Helped |
| Add the confidence loss using SVHN as training OOD data | VGGNet trained on CIFAR-10 · CIFAR-10 vs SVHN | TNR at TPR 95%: 13.7 -> 99.8 | Helped |
| Add the confidence loss using SVHN as training OOD data | VGGNet trained on CIFAR-10 · CIFAR-10 vs TinyImageNet, LSUN, and Gaussian | TNR at TPR 95%: 13.6 / 14.0 / 2.8 -> 9.9 / 10.5 / 3.3 | Mixed |
| Train with the joint confidence loss instead of cross entropy | VGGNet · CIFAR-10 | classification accuracy: 80.14 -> 81.39 | Helped |
| Train with the joint confidence loss instead of cross entropy | VGGNet · SVHN | classification accuracy: 93.82 -> 93.81 | No clear effect |
| Train with the joint confidence loss instead of cross entropy | VGGNet · CIFAR-10 | ECE: 18.45 -> 14.62 | Helped |
| Train with the joint confidence loss instead of cross entropy | VGGNet · SVHN | ECE: 5.30 -> 5.13 | Helped |
| Apply temperature scaling to the joint confidence-loss model rather than the temperature-scaled cross-entropy model | VGGNet · CIFAR-10 | ECE: 7.07 -> 6.19 | Helped |
| Apply temperature scaling to the joint confidence-loss model rather than the temperature-scaled cross-entropy model | VGGNet · SVHN | ECE: 2.80 -> 1.39 | Helped |
| Use the proposed GAN instead of the original GAN to generate training OOD samples (ablation) | Confidence-loss classifier · OOD detection across tested in- and out-of-distribution pairs | detection performance | Helped |
| Jointly train the classifier and proposed GAN instead of using explicit OOD training data (ablation) | Confidence-loss classifier · OOD detection across tested in- and out-of-distribution pairs | detection performance | Helped |
| Use samples from an original GAN rather than the proposed GAN for the confidence loss (ablation) | Confidence-loss classifier · OOD detection across tested in- and out-of-distribution pairs | detection performance | Hurt |
Closest papers by meaning
- Out-of-Distribution Detection with Prototypical Outlier Proxy (AAAI 2025 · similarity 0.86)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.86)
- CASE: Exploiting Intra-class Compactness and Inter-class Separability of Feature Embeddings for Out-of-Distribution Detection (AAAI 2024 · similarity 0.86)
- DOS: Diverse Outlier Sampling for Out-of-Distribution Detection (ICLR 2024 · similarity 0.85)
- Out-of-Distribution Detection in Long-Tailed Recognition with Calibrated Outlier Class Learning (AAAI 2024 · similarity 0.85)
- ProHOC: Probabilistic Hierarchical Out-of-Distribution Classification via Multi-Depth Networks (CVPR 2025 · similarity 0.85)
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.