Research map / Large language models / Alignment and preferences
Nonparametric LLM Evaluation from Preference Data
Dennis Frauen, Athiya Deviyani, Mihaela van der Schaar, Stefan Feuerriegel
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 debiased estimator instead of plug-in | Borda score, synthetic · n=1000 | estimation error: 0.38 -> 0.15 | Improved |
| use debiased estimator instead of plug-in | BT score, synthetic · n=1000 | estimation error: 0.62 -> 0.25 | Improved |
| use debiased estimator instead of plug-in | Rank centrality, synthetic · n=3000 | estimation error: 0.18 -> 0.09 | Improved |
| use debiased estimator instead of plug-in | Borda score, synthetic · n=1000 | CI coverage: 0.17 -> 0.94 | Improved |
| use debiased estimator instead of plug-in | BT score, synthetic · n=2000 | CI coverage: 0.85 vs 0.07 | Mixed |
| use A-optimal labeling policy instead of random | Borda estimator · budget 2000 | ranking MSE: 0.141 -> 0.130 | Improved |
| use A-optimal labeling policy instead of random | BT estimator · budget 2000 | ranking MSE: 2.974 -> 2.861 | Improved |
| use A-optimal labeling policy instead of random | Rank centrality estimator · budget 2000 | ranking MSE: 0.020 -> 0.017 | Improved |
| add higher quality external judges as features | debiased estimators · synthetic judges | relative ranking error | Improved |
| use debiased BT projection instead of restricted debiased estimator under BT misspecification | DMLRank BT · BT misspecification | estimation error | Improved |
Closest papers by meaning
- Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference (ICML 2024 · similarity 0.89)
- A Statistical Framework for Ranking LLM-based Chatbots (ICLR 2025 · similarity 0.88)
- Prompt-to-Leaderboard: Prompt-Adaptive LLM Evaluations (ICML 2025 · similarity 0.88)
- Bridging Human and LLM Judgments: Understanding and Narrowing the Gap (NeurIPS 2025 · similarity 0.88)
- Preference Learning Algorithms Do Not Learn Preference Rankings (NeurIPS 2024 · similarity 0.88)
- From Uncertain Judgments to Calibrated Rankings: Conformal Elo Estimation for LLM Evaluation (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.
