Research map / Learning theory and algorithms / Prediction and decision losses
To Trust Or Not To Trust A Classifier
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 trust score instead of the classifier's confidence score to rank predictions | Pretrained classifiers on UCI datasets · Five UCI benchmark datasets | precision at recall percentile | Helped |
| Use the trust score instead of the classifier's confidence score to rank predictions | Neural networks, random forests, and logistic regression · UCI datasets | precision at recall percentile | Helped |
| Use the trust score instead of classifier confidence to identify trustworthy or suspicious examples | CNN classifiers · MNIST, SVHN, CIFAR-10, and CIFAR-100 | precision at recall percentile | Mixed |
| Increase the classifier's training iterations | Neural network on Digits · Digits | precision at recall percentile | Mixed |
| Compute trust scores using intermediate network representations rather than raw inputs (ablation) | CNN classifiers · MNIST, SVHN, CIFAR-10, and CIFAR-100 | precision at recall percentile | No clear effect |
| Select a nonzero density-filtering fraction using cross-validation instead of using no filtering (ablation) | Trust score on large image datasets · MNIST, SVHN, CIFAR-10, and CIFAR-100 | precision at recall percentile | No clear effect |
| Skip the initial density-filtering step (ablation) | Trust score on large image datasets · MNIST, SVHN, CIFAR-10, and CIFAR-100 | precision at recall percentile | No clear effect |
| Reduce intermediate representations to 20 dimensions using PCA before computing trust scores (ablation) | Trust score on large image datasets · MNIST, SVHN, CIFAR-10, and CIFAR-100 | precision at recall percentile | No clear effect |
Closest papers by meaning
- Naive Bayes Classifiers over Missing Data: Decision and Poisoning (ICML 2024 · similarity 0.82)
- Trust Regions for Explanations via Black-Box Probabilistic Certification (ICML 2024 · similarity 0.82)
- U-trustworthy Models. Reliability, Competence, and Confidence in Decision-Making (AAAI 2024 · similarity 0.82)
- Navigating Conflicting Views: Harnessing Trust for Learning (ICML 2025 · similarity 0.82)
- Trustworthy Machine Learning through Data-Specific Indistinguishability (ICML 2025 · similarity 0.81)
- Data-Driven Discovery of Design Specifications (Student Abstract) (AAAI 2024 · similarity 0.81)
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.