arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2605.09702stat.MEcs.CL
收藏

校准,而非筛选:从噪声LLM评判者中高效估计

Calibrate, Don't Curate: Label-Efficient Estimation from Noisy LLM Judges

Department of Biostatistics(生物统计学系)

Yanran Li

AI总结:

本文研究了在噪声LLM评判者中如何通过校准而非筛选来提高估计效率,发现校准全组评判者优于基于准确率的筛选,且在多个基准测试中表现更优。

摘要:

多评判者评估日益被用于评估LLM和奖励模型,传统方法是筛选:保留最准确的评判者并丢弃较弱的。我们证明当目标不是点准确度,而是从标记的校准集获得校准的概率评估时,此方法会反转。在保持聚合和校准过程不变的情况下,我们比较了按准确性排名的前k个评判者选择与使用完整评判者小组。在四个涵盖LLM作为评判者和奖励模型设置的标记成对评估基准上,校准的完整小组始终优于基于准确性的选择。在RewardBench2上,保留所有评判者达到负对数似然(NLL)为0.006,而前5个选择则为0.013,校准误差减半。这一优势在评判者家庭去重和更强的同流程子集搜索后仍持续存在。我们通过 oracle 分析解释了这种反转,显示在适当评分规则下,最优校准风险在获得额外评判者信号时不会增加,且即使低于机会率的评判者在它们的偏见可学习且信号非冗余时也可能是有用的。由此得出的操作原则是简单的:在具有标记校准数据的多评判者评估中,不要仅凭准确性丢弃弱评判者;在它们可解析、非冗余且可校准时保留它们。

英文摘要:

Multi-judge evaluation is increasingly used to assess LLMs and reward models, and the prevailing heuristic is to curate: keep the most accurate judges and discard weaker ones. We show that this heuristic can reverse when the target is not point accuracy, but calibrated probabilistic evaluation from a labeled calibration set. Holding the aggregation and calibration procedures fixed, we compare accuracy-ranked top-$k$ judge selection with using the full judge panel. Across four labeled pairwise-evaluation benchmarks spanning LLM-as-judge and reward-model settings, the calibrated full panel consistently outperforms accuracy-based selection. On RewardBench2, retaining all judges achieves negative log-likelihood (NLL) of $0.006$ versus $0.013$ under top-5 selection, halving the calibration error. This advantage persists after judge-family deduplication and against stronger same-pipeline subset search. We explain this reversal with oracle analyses showing that the optimal calibrated risk under proper scoring rules cannot increase when additional judge signals are made available, and that even below-chance judges can be useful when their biases are learnable and their signals are non-redundant. The resulting operating principle is simple: in multi-judge evaluation with labeled calibration data, do not discard weak judges by accuracy alone; keep them when they are parseable, non-redundant, and calibratable.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑