Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Spectral Conformal Risk Control: Distribution-Free Tail Guarantees via Bayesian Quadrature
Mohammad Mahdi Kazemi Esfeh, Qi Yan, Yongxing Zhang, Zahra Gholami, Renjie Liao, Purang Abolmaesumi
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 mean-risk control (CRC) with BQ-SRC tail-focused spectral risk control | BQ-SRC · Heteroskedastic regression | violation rate (%): 46.18% -> 0.90% | Improved |
| replace mean-risk control (CRC) with BQ-SRC | BQ-SRC · Synthetic n=10 | violation rate (%): 21.46% -> 0.00% | Improved |
| use binomial lower confidence bound instead of DKW inflation (ablation) | BQ-SRC · Monte Carlo conservatism M=5000 | quantile margin: 0.0276 -> 0.0096 | Improved |
| use BQ-SRC with binomial LCB instead of RCPS | MS-COCO multilabel · MS-COCO | prediction-set size: 3.57 -> 3.39 | Improved |
| add two-parameter (tau, k) control (ablation) | BQ-SRC-LCB · MS-COCO | violation rate (%) and set size: violations 1.80% -> 0.60%, set size 3.39 -> 3.47 | Mixed |
| use BQ-SRC (mix 95/5) instead of RCPS in regression | BQ-SRC · Heteroskedastic regression | interval length: 14.32 -> 15.62 | Hurt |
| use mixture 95/5 spectrum instead of mean spectrum (ablation) | BQ-SRC · ADE20K segmentation APS | empirical risk: 0.0984 -> 0.0824 | Improved |
| use pure CVaR 0.9 spectrum for segmentation (ablation) | BQ-SRC · ADE20K segmentation APS | mean set size: 1.668 -> 134.062 | Hurt |
| use mean spectrum within BQ-SRC (ablation) | BQ-SRC · ADE20K segmentation | empirical risk: 0.0989 -> 0.0984 | No clear effect |
| increase Monte Carlo budget beyond 2,000 samples (ablation) | BQ-SRC · Monte Carlo budget ablation | violation rate | No clear effect |
| use BQ-SRC (mix 95/5) instead of BQC | BQ-SRC · Synthetic n=200 | violation rate (%): 3.00% -> 0.80% | Improved |
Closest papers by meaning
- Conformal Risk Control (ICLR 2024 · similarity 0.87)
- Conformal Risk Training: End-to-End Optimization of Conformal Risk Control (NeurIPS 2025 · similarity 0.85)
- Conformal Prediction as Bayesian Quadrature (ICML 2025 · similarity 0.84)
- Dual Risk Minimization: Towards Next-Level Robustness in Fine-tuning Zero-Shot Models (NeurIPS 2024 · similarity 0.84)
- Learning Polyhedral Conformal Sets for Robust Optimization (NeurIPS 2026 · similarity 0.84)
- From Risk to Uncertainty: Generating Predictive Uncertainty Measures via Bayesian Estimation (ICLR 2025 · similarity 0.83)
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.
