Research map / Label-efficient and robust learning / Recognition and long-tail
Enhancing Conformal Prediction via Class Similarity
Ariel Fargion, Lahav Dabah, Tom Tirer
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add model-agnostic out-of-group penalty (MA-CS) | RAPS · CIFAR-100 ResNet34, alpha=0.05 | average set size: 5.79 -> 4.52 | Improved |
| add model-specific similarity penalty (MS-CS) | RAPS · CIFAR-100 ResNet34, alpha=0.05 | average set size: 5.79 -> 3.79 | Improved |
| add model-agnostic penalty (MA-CS) | RAPS · CIFAR-100 ResNet34, alpha=0.05 | #superclasses in set: 3.40 -> 2.29 | Improved |
| add model-specific penalty (MS-CS) | LAC · CIFAR-100 ResNet50, alpha=0.05 | average set size: 3.68 -> 2.92 | Improved |
| add model-specific penalty (MS-CS) | SAPS · CIFAR-100 ResNet50, alpha=0.05 | average set size: 3.45 -> 3.14 | Improved |
| use MS-CS instead of MA-CS | CP with class-similarity penalty · CIFAR-100 ResNet34, RAPS, alpha=0.05 | average set size: 4.52 -> 3.79 | Improved |
| replace score-based class clustering (Clustered CP) vs standard | LAC/RAPS/SAPS · CIFAR-100 ResNet50, alpha=0.1 | average set size: 1.62 -> 1.82 (LAC) | Hurt |
| use hierarchical AIR baseline instead of standard score | RAPS · CIFAR-100 ResNet34, alpha=0.05 | average set size: 5.79 -> 15.15 | Hurt |
| use AIR instead of standard score | RAPS · CIFAR-100 ResNet34, alpha=0.05 | #superclasses in set: 3.40 -> 3.03 | Improved |
| increase penalty lambda from zero to small values (ablation) | RAPS + MA-CS · CIFAR-100 ResNet50, alpha=0.1 | average set size | Improved |
| increase penalty lambda to large values (ablation) | RAPS + MA-CS · CIFAR-100 ResNet50, alpha=0.1 | average set size | Hurt |
| vary lambda (ablation) | RAPS + MA-CS · CIFAR-100 ResNet50, alpha=0.1 | worst class-conditional coverage gap | No clear effect |
Closest papers by meaning
- Length Optimization in Conformal Prediction (NeurIPS 2024 · similarity 0.88)
- Conformal Prediction for Class-wise Coverage via Augmented Label Rank Calibration (NeurIPS 2024 · similarity 0.88)
- Conformity Score Averaging for Classification (ICML 2025 · similarity 0.87)
- Class Adaptive Conformal Training (NeurIPS 2026 · similarity 0.87)
- Hierarchical Conformal Classification (NeurIPS 2026 · similarity 0.87)
- On Temperature Scaling and Conformal Prediction of Deep Classifiers (ICML 2025 · 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.
