Research map / Label-efficient and robust learning / OOD and anomaly detection
A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks
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 |
|---|---|---|---|
| Use the maximum predicted-class softmax probability as an error-detection score | Image, text, and speech classifiers · Error detection across reported classification tasks | AUROC: 97 | Helped |
| Use maximum softmax probability for out-of-distribution detection | Image classifiers · CIFAR-10/SUN | AUROC: 95 | Helped |
| Use maximum softmax probability for out-of-distribution detection | Text classifiers · 15/5 Newsgroups | AUROC: 75 | Helped |
| Use maximum softmax probability for out-of-distribution detection | WSJ part-of-speech tagger · WSJ/Weblog | AUROC: 61 | No clear effect |
| Add an auxiliary decoder and abnormality module trained on clean and noised examples | TIMIT classifier · TIMIT/+Airport | AUROC: 75 -> 100 | Helped |
| Add an auxiliary decoder and abnormality module trained on clean and noised examples | TIMIT classifier · TIMIT/Chinese | AUROC: 79 -> 90 | Helped |
| Add an auxiliary decoder and abnormality module trained on clean and noised examples | MNIST classifier · MNIST/Omniglot | AUROC: 95 -> 100 | Helped |
| Add an auxiliary decoder and abnormality module trained on clean and noised examples | MNIST classifier · MNIST/Gaussian | AUROC: 88 -> 100 | Helped |
| Ignore the blank-symbol logit when computing speech softmax detection scores | TIMIT phone recognizer · TIMIT out-of-distribution speech | AUROC | Helped |
| Use KL divergence from the uniform softmax distribution instead of maximum softmax probability | Neural network classifiers · Error and out-of-distribution detection | AUROC | No clear effect |
Closest papers by meaning
- Out-of-Distribution Detection using Multiple Semantic Label Representations (NeurIPS 2018 · similarity 0.87)
- A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks (NeurIPS 2018 · similarity 0.86)
- Detecting Out-of-Distribution Examples with Gram Matrices (ICML 2020 · similarity 0.86)
- Subclass-Dominant Label Noise: A Counterexample for the Success of Early Stopping (NeurIPS 2023 · similarity 0.86)
- Sparsity as a Key: Unlocking New Insights from Latent Structures for Out-of-Distribution Detection (CVPR 2026 · similarity 0.86)
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.84)
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.