Research map / Label-efficient and robust learning / Recognition and long-tail
Ensuring Class-Conditional Coverage for Pathological Workflows (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Applies Classwise (Mondrian) conformal prediction with LAC scores to CRIC and BreakHis cancer datasets to achieve class-conditional coverage, reducing the average class coverage gap versus baseline split CP.
Problem
Marginal coverage in conformal prediction can give inconsistent per-class coverage, which is risky for high-stakes computational pathology workflows where rare but serious diagnoses must be reliably included in prediction sets.
Approach
Calibrates conformal thresholds separately within each class (Mondrian/Classwise CP) so every class meets the user-specified coverage level, and measures the resulting class-conditional coverage gap on cancer datasets.
Key results
- Classwise CP reduces CovGap to 0.0 at 90% and 95% coverage on CRIC and BreakHis, versus 4.59 and 2.44 for Baseline on CRIC.
- At 99% coverage, Classwise CovGap is 0.79 on CRIC and 0.66 on BreakHis, versus 1.09 and 1.30 for Baseline.
- Reported reductions in CovGap range from 27.52% to 100% across datasets and coverage levels.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use classwise calibration instead of pooled calibration | CP with DieT · CRIC, Cov = 90% | CovGap: 4.59 (4.16) -> 0.0 (0.0) | Helped |
| Use classwise calibration instead of pooled calibration | CP with DieT · CRIC, Cov = 95% | CovGap: 2.44 (2.30) -> 0.0 (0.0) | Helped |
| Use classwise calibration instead of pooled calibration | CP with DieT · CRIC, Cov = 99% | CovGap: 1.09 (1.18) -> 0.79 (0.99) | Helped |
| Use classwise calibration instead of pooled calibration | CP with DieT · BreakHis, Cov = 90% | CovGap: 4.44 (4.07) -> 0.0 (0.0) | Helped |
| Use classwise calibration instead of pooled calibration | CP with DieT · BreakHis, Cov = 95% | CovGap: 1.82 (1.88) -> 0.0 (0.0) | Helped |
| Use classwise calibration instead of pooled calibration | CP with DieT · BreakHis, Cov = 99% | CovGap: 1.30 (1.41) -> 0.66 (0.92) | Helped |
Limitations
- Evaluation limited to two cancer datasets and a single underlying model (DieT).
- Abstract-length study with no code or detailed hardware reported.
- Coverage gap is measured only empirically without theoretical analysis of finite-sample guarantees under class imbalance.
Experimental setup
Datasets: CRIC BreakHis
Benchmarks: CRIC BreakHis
Metrics: average class coverage gap (CovGap)
Baselines: Baseline split conformal prediction (marginal coverage)
Builds on: LAC conformal score Mondrian conformal prediction split conformal prediction
Closest papers by meaning
- Conformal Prediction for Class-wise Coverage via Augmented Label Rank Calibration (NeurIPS 2024 · similarity 0.90)
- Reliable Uncertainty Quantification in Machine Learning via Conformal Prediction (AAAI 2025 · similarity 0.87)
- Kandinsky Conformal Prediction: Beyond Class- and Covariate-Conditional Coverage (ICML 2025 · similarity 0.87)
- Backward Conformal Prediction (NeurIPS 2025 · similarity 0.86)
- Robust Conditional Conformal Prediction via Branched Normalizing Flow (NeurIPS 2026 · similarity 0.86)
- Probabilistic Conformal Prediction with Approximate Conditional Validity (ICLR 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.