Research map / Label-efficient and robust learning / Recognition and long-tail
Conformal Prediction for Long-Tailed Classification
Tiffany Ding, Jean-Baptiste Fermanian, Joseph Salmon
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 |
|---|---|---|---|
| replace softmax score with prevalence-adjusted softmax (PAS) in Standard CP | Standard CP · Pl@ntNet-300K (alpha=0.1) | classes with coverage below 50%: 421 -> 180 classes (set size 1.57 -> 2.57) | Improved |
| replace softmax score with PAS in Standard CP | Standard CP · Pl@ntNet-300K (alpha=0.1) | average set size: 1.57 -> 2.57 | Hurt |
| use Standard with PAS instead of Classwise | Standard CP · Pl@ntNet-300K (alpha=0.1) | average set size: 780 -> 2.57 | Improved |
| interpolate Classwise and Standard quantiles with tau=0.99 instead of tau=1 (ablation) | Interp-Q · Pl@ntNet-300K (alpha=0.1) | average set size: 780 -> 7.6 | Improved |
| interpolate Classwise and Standard quantiles with tau=0.99 instead of tau=1 (ablation) | Interp-Q · iNaturalist-2018 (alpha=0.1) | average set size: 7430 -> 55.8 | Improved |
| upweight at-risk species by lambda in WPAS score (ablation) | Standard with WPAS · Pl@ntNet-300K at-risk species | class-conditional coverage of at-risk classes | Improved |
| add PAS score to Standard CP for human decision accuracy | Standard with PAS · truncated Pl@ntNet and iNaturalist-2018 simulated decision-makers | worst-class decision accuracy | Improved |
| tune alpha in Standard CP to target class-conditional coverage instead of using PAS/Interp-Q | Standard CP · Pl@ntNet-300K and iNaturalist-2018 | set size vs class-conditional coverage trade-off | Hurt |
Closest papers by meaning
- Class conditional conformal prediction for multiple inputs by p-value aggregation (NeurIPS 2025 · similarity 0.89)
- Conformal Prediction for Class-wise Coverage via Augmented Label Rank Calibration (NeurIPS 2024 · similarity 0.88)
- Kandinsky Conformal Prediction: Beyond Class- and Covariate-Conditional Coverage (ICML 2025 · similarity 0.88)
- Ensuring Class-Conditional Coverage for Pathological Workflows (Student Abstract) (AAAI 2025 · similarity 0.87)
- Backward Conformal Prediction (NeurIPS 2025 · similarity 0.86)
- Conformity Score Averaging for Classification (ICML 2025 · similarity 0.86)
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.
