Research map / Large language models / Language, safety and interpretability
Multi-LLM Adaptive Conformal Inference for Reliable LLM Response
Kangjun Noh, Seongchan Lee, Ilmun Kim, Kyungwoo Song
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 |
|---|---|---|---|
| use MACI multiplicative group-conditional conformal filtering instead of BCI | MACI · MedLFQA, 90% target | retention ratio: 0.02 -> 0.50 | Improved |
| use MACI instead of CCI | MACI · WikiBio, 90% target | retention ratio: 0.11 -> 0.25 | Improved |
| use MACI instead of CCI on ExpertQA at high coverage targets | MACI · ExpertQA, 95% target | retention ratio: 0.17 -> 0.10 | Hurt |
| group-conditional calibration (Mondrian) in place of a single global threshold (ablation) | MACI · MedLFQA/WikiBio/ExpertQA groups | group coverage | Improved |
| replace worst-claim conformity score with cumulative product of factuality-scores (ablation) | MACI · all datasets | retention ratio | Improved |
| ensemble three LLM scorers with FPR-optimized weights instead of single LLM or arithmetic mean (ablation) | MACI factuality-score · MedLFQA | FPR, MSE and retention | Improved |
| add density-ratio resampling of calibration set (MACI-DRE) under covariate shift (ablation) | MACI · MedLFQA covariate shift, Medium risk, 90% target | coverage: 0.82 -> 0.94 | Mixed |
| use MACI single-pass scoring and streamlined calibration instead of CCI | MACI · WikiBio, 500 test samples | wall-clock time (s): 1643.91 -> 598.98 | Improved |
| let CCI use adaptive alpha to match MACI retention | CCI · WikiBio | coverage | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SelfCheck — SelfCheck: Using LLMs to Zero-Shot Check Their Own Step-by-Step Reasoning (ICLR 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Large language model validity via enhanced conformal prediction methods (NeurIPS 2024 · similarity 0.91)
- Language Models with Conformal Factuality Guarantees (ICML 2024 · similarity 0.88)
- Long-form factuality in large language models (NeurIPS 2024 · similarity 0.88)
- Prune 'n Predict: Optimizing LLM Decision-making with Conformal Prediction (ICML 2025 · similarity 0.87)
- Sample Smart, Not Hard: Correctness-First Decoding for Better Reasoning in LLMs (ICLR 2026 · similarity 0.87)
- Statistical Hypothesis Testing for Auditing Robustness in Language Models (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.
