Research map / Label-efficient and robust learning / Recognition and long-tail
Class-Conditional Conformal Prediction with Many Classes
Tiffany Ding, Anastasios Nikolas Angelopoulos, Stephen Bates, Michael Jordan, Ryan Tibshirani
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 |
|---|---|---|---|
| Cluster classes with similar score distributions and calibrate at the cluster level instead of using one global or one per-class threshold | Conformal prediction with softmax scores · ImageNet, n_avg=20 | CovGap: 5.2 (0.0) -> 3.9 (0.0) | Improved |
| Cluster classes with similar score distributions and calibrate at the cluster level instead of using one global or one per-class threshold | Conformal prediction with softmax scores · CIFAR-100, n_avg=20 | CovGap: 4.0 (0.0) -> 3.6 (0.1) | Improved |
| Cluster classes with similar score distributions and calibrate at the cluster level instead of using one global or one per-class threshold | Conformal prediction with softmax scores · Places365, n_avg=20 | CovGap: 4.5 (0.0) -> 3.9 (0.1) | Improved |
| Cluster classes with similar score distributions and calibrate at the cluster level instead of using one global or one per-class threshold | Conformal prediction with softmax scores · iNaturalist, n_avg=20 | CovGap: 7.0 (0.1) -> 6.4 (0.1) | Improved |
| Use classwise rather than clustered calibration | Conformal prediction with softmax scores · ImageNet, n_avg=20 | AvgSize: 2.7 (0.1) -> 23.0 (0.9) | Hurt |
| Use classwise rather than clustered calibration | Conformal prediction with softmax scores · iNaturalist, n_avg=20 | AvgSize: 3.4 (0.1) -> 364.0 (1.4) | Hurt |
| Increase the average calibration examples per class | Classwise conformal · ImageNet, softmax | CovGap: 7.7 (0.0) -> 2.8 (0.0) | Improved |
| Increase the average calibration examples per class | Standard conformal · ImageNet, softmax | CovGap: 5.2 (0.0) -> 5.2 (0.0) | No clear effect |
| Use APS rather than softmax scores | Standard conformal · ImageNet | CovGap: 5.2 (0.0) -> 2.6 (0.0) | Improved |
| Use RAPS rather than APS scores | Standard conformal · ImageNet | AvgSize: 25.8 (0.1) -> 5.1 (0.0) | Improved |
| Randomize the conformal quantile to target exact rather than at-least nominal coverage | Classwise conformal · All evaluated datasets and scores | CovGap | Improved |
| Randomize the conformal quantile to target exact rather than at-least nominal coverage | Standard and clustered conformal · All evaluated datasets and scores | CovGap | No clear effect |
Closest papers by meaning
- Conformal Prediction for Class-wise Coverage via Augmented Label Rank Calibration (NeurIPS 2024 · similarity 0.89)
- Kandinsky Conformal Prediction: Beyond Class- and Covariate-Conditional Coverage (ICML 2025 · similarity 0.89)
- Ensuring Class-Conditional Coverage for Pathological Workflows (Student Abstract) (AAAI 2025 · similarity 0.89)
- Conformal Prediction with Learned Features (ICML 2024 · similarity 0.88)
- Class conditional conformal prediction for multiple inputs by p-value aggregation (NeurIPS 2025 · similarity 0.88)
- Hierarchical Conformal Classification (NeurIPS 2026 · similarity 0.88)
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.
