Efficient Inference for Noisy LLM-as-a-Judge Evaluation

• URL: http://arxiv.org/abs/2601.05420v1
• Date: Thu, 08 Jan 2026 22:46:26 GMT
• Title: Efficient Inference for Noisy LLM-as-a-Judge Evaluation
• Authors: Yiqun T Chen, Sizhu Lu, Sijia Li, Moran Guo, Shengyi Li,
• Abstract summary: Large language models (LLMs) are increasingly used as automatic evaluators of generative AI outputs.<n>In practice, LLM judges are imperfect predictions for the underlying truth and can exhibit systematic, non-random errors.
• Score: 8.2511120576505
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models (LLMs) are increasingly used as automatic evaluators of generative AI outputs, a paradigm often referred to as "LLM-as-a-judge." In practice, LLM judges are imperfect predictions for the underlying truth and can exhibit systematic, non-random errors. Two main approaches have recently been proposed to address this issue: (i) direct measurementerror correction based on misclassification models such as Rogan-Gladen-style estimators, and (ii) surrogate-outcome approaches such as prediction-powered inference (PPI), which correct bias by calibrating prediction residuals on a small set of gold-standard human labels. In this paper, we systematically study the performance of these two approaches for estimating mean parameters (e.g., average benchmark scores or pairwise win rates). Leveraging tools from semiparametric efficiency theory, we unify the two classes of estimators by deriving explicit forms of efficient influence function (EIF)-based efficient estimators and characterize conditions under which PPI-style estimators attain strictly smaller asymptotic variance than measurement-error corrections. We verify our theoretical results in simulations and demonstrate the methods on real-data examples. We provide an implementation of the benchmarked methods and comparison utilities at https://github.com/yiqunchen/debias-llm-as-a-judge.

Related papers

• Evaluating LLMs When They Do Not Know the Answer: Statistical Evaluation of Mathematical Reasoning via Comparative Signals [18.612081365101464]
  We develop a framework that combines standard labeled outcomes with pairwise comparison signals obtained by having models judge auxiliary reasoning chains.<n>Across simulations, our one-step estimator substantially improves ranking accuracy with gains increasing as model output noise grows.<n>Experiments on GPQA Diamond, AIME 2025 and GSM8K further demonstrate more precise performance estimation and more reliable model rankings.
  arXiv  Detail & Related papers  (2026-02-03T03:40:01Z)
• Efficient Inference Using Large Language Models with Limited Human Data: Fine-Tuning then Rectification [2.503562746177713]
  We develop a framework that combines fine-tuning and rectification, and optimally allocates limited labeled samples across the two stages.<n>Building on this insight, we leverage empirical scaling laws to develop a data-driven method for optimally splitting samples between the fine-tuning and rectification stages.<n> Empirical analysis validates our framework, demonstrating improved estimation and inference performance compared to using either fine-tuning or rectification alone.
  arXiv  Detail & Related papers  (2025-11-23T05:23:21Z)
• Self-Boost via Optimal Retraining: An Analysis via Approximate Message Passing [58.52119063742121]
  Retraining a model using its own predictions together with the original, potentially noisy labels is a well-known strategy for improving the model performance.<n>This paper addresses the question of how to optimally combine the model's predictions and the provided labels.<n>Our main contribution is the derivation of the Bayes optimal aggregator function to combine the current model's predictions and the given labels.
  arXiv  Detail & Related papers  (2025-05-21T07:16:44Z)
• Statistical Inference for Generative Model Comparison [8.366326412272898]
  We develop a method for comparing how close different generative models are to the underlying distribution of test samples.<n>Our approach employs the Kullback-Leibler (KL) divergence to measure the distance between a generative model and the unknown test distribution.<n>We extend our method to comparing conditional generative models and leverage Edgeworth expansions to address limited-data settings.
  arXiv  Detail & Related papers  (2025-01-31T05:31:05Z)
• Achieving $\widetilde{\mathcal{O}}(\sqrt{T})$ Regret in Average-Reward POMDPs with Known Observation Models [69.1820058966619]
  We tackle average-reward infinite-horizon POMDPs with an unknown transition model.<n>We present a novel and simple estimator that overcomes this barrier.
  arXiv  Detail & Related papers  (2025-01-30T22:29:41Z)
• CalibraEval: Calibrating Prediction Distribution to Mitigate Selection Bias in LLMs-as-Judges [21.580762639442913]
  We introduce CalibraEval, a novel label-free method for mitigating selection bias during inference. CalibraEval reformulates debiasing as an optimization task aimed at adjusting observed prediction distributions to align with unbiased prediction distributions. We show that CalibraEval effectively mitigates selection bias and improves performance compared to existing debiasing methods.
  arXiv  Detail & Related papers  (2024-10-20T13:47:39Z)
• Weak Supervision Performance Evaluation via Partial Identification [46.73061437177238]
  Programmatic Weak Supervision (PWS) enables supervised model training without direct access to ground truth labels. We present a novel method to address this challenge by framing model evaluation as a partial identification problem. Our approach derives reliable bounds on key metrics without requiring labeled data, overcoming core limitations in current weak supervision evaluation techniques.
  arXiv  Detail & Related papers  (2023-12-07T07:15:11Z)
• Towards Calibrated Robust Fine-Tuning of Vision-Language Models [97.19901765814431]
  This work proposes a robust fine-tuning method that improves both OOD accuracy and confidence calibration simultaneously in vision language models. We show that both OOD classification and OOD calibration errors have a shared upper bound consisting of two terms of ID data. Based on this insight, we design a novel framework that conducts fine-tuning with a constrained multimodal contrastive loss enforcing a larger smallest singular value.
  arXiv  Detail & Related papers  (2023-11-03T05:41:25Z)
• Leveraging Variational Autoencoders for Parameterized MMSE Estimation [10.141454378473972]
  We propose a variational autoencoder-based framework for parameterizing a conditional linear minimum mean squared error estimator. The derived estimator is shown to approximate the minimum mean squared error estimator by utilizing the variational autoencoder as a generative prior for the estimation problem. We conduct a rigorous analysis by bounding the difference between the proposed and the minimum mean squared error estimator.
  arXiv  Detail & Related papers  (2023-07-11T15:41:34Z)
• The Challenges of Hyperparameter Tuning for Accurate Causal Effect Estimation [2.43420394129881]
  Many ML methods (causal estimators') have been proposed for causal inference.<n>For non-causal predictive tasks, there is a consensus on the choice of tuning metrics, making it simple to compare models.<n>For causal inference tasks, such a consensus is yet to be reached, making any comparison of causal models difficult.
  arXiv  Detail & Related papers  (2023-03-02T17:03:02Z)
• Leveraging Unlabeled Data to Predict Out-of-Distribution Performance [63.740181251997306]
  Real-world machine learning deployments are characterized by mismatches between the source (training) and target (test) distributions. In this work, we investigate methods for predicting the target domain accuracy using only labeled source data and unlabeled target data. We propose Average Thresholded Confidence (ATC), a practical method that learns a threshold on the model's confidence, predicting accuracy as the fraction of unlabeled examples.
  arXiv  Detail & Related papers  (2022-01-11T23:01:12Z)
• Learning to Estimate Without Bias [57.82628598276623]
  Gauss theorem states that the weighted least squares estimator is a linear minimum variance unbiased estimation (MVUE) in linear models. In this paper, we take a first step towards extending this result to non linear settings via deep learning with bias constraints. A second motivation to BCE is in applications where multiple estimates of the same unknown are averaged for improved performance.
  arXiv  Detail & Related papers  (2021-10-24T10:23:51Z)
• Scalable Personalised Item Ranking through Parametric Density Estimation [53.44830012414444]
  Learning from implicit feedback is challenging because of the difficult nature of the one-class problem. Most conventional methods use a pairwise ranking approach and negative samplers to cope with the one-class problem. We propose a learning-to-rank approach, which achieves convergence speed comparable to the pointwise counterpart.
  arXiv  Detail & Related papers  (2021-05-11T03:38:16Z)
• Doubly Robust Semiparametric Difference-in-Differences Estimators with High-Dimensional Data [15.27393561231633]
  We propose a doubly robust two-stage semiparametric difference-in-difference estimator for estimating heterogeneous treatment effects. The first stage allows a general set of machine learning methods to be used to estimate the propensity score. In the second stage, we derive the rates of convergence for both the parametric parameter and the unknown function.
  arXiv  Detail & Related papers  (2020-09-07T15:14:29Z)
• Machine learning for causal inference: on the use of cross-fit estimators [77.34726150561087]
  Doubly-robust cross-fit estimators have been proposed to yield better statistical properties. We conducted a simulation study to assess the performance of several estimators for the average causal effect (ACE) When used with machine learning, the doubly-robust cross-fit estimators substantially outperformed all of the other estimators in terms of bias, variance, and confidence interval coverage.
  arXiv  Detail & Related papers  (2020-04-21T23:09:55Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.