Efficient Model Performance Evaluation Using a Combination of Expert and Crowd-sourced Labels

Sam Corbett-Davies, Viet-An Nguyen, Udi Weinsberg
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:4510-4518, 2026.

Abstract

As models, particularly large language models (LLMs), are deployed on increasingly challenging tasks, correctly evaluating their performance is growing in importance and difficulty. Expert human labelers are high-quality but scarce and resource-intensive to obtain, while crowd-sourced labels are more readily accessible at scale but lower in quality. We propose Maven (Model And Voter EvaluatioN), a hierarchical Bayesian model that combines these two label sources to produce model performance estimates on binary tasks that are less biased than using crowd-sourced labels alone and have lower variance than using expert labels alone. By modeling the ranking of model scores, Maven is robust to a range of prediction distributions and achieves constant inference time regardless of dataset size. We validate our approach on both simulated and real-world data, and deploy it to measure production models at Meta.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-corbett-davies26a, title = { Efficient Model Performance Evaluation Using a Combination of Expert and Crowd-sourced Labels }, author = {Corbett-Davies, Sam and Nguyen, Viet-An and Weinsberg, Udi}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {4510--4518}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/corbett-davies26a/corbett-davies26a.pdf}, url = {https://proceedings.mlr.press/v300/corbett-davies26a.html}, abstract = { As models, particularly large language models (LLMs), are deployed on increasingly challenging tasks, correctly evaluating their performance is growing in importance and difficulty. Expert human labelers are high-quality but scarce and resource-intensive to obtain, while crowd-sourced labels are more readily accessible at scale but lower in quality. We propose Maven (Model And Voter EvaluatioN), a hierarchical Bayesian model that combines these two label sources to produce model performance estimates on binary tasks that are less biased than using crowd-sourced labels alone and have lower variance than using expert labels alone. By modeling the ranking of model scores, Maven is robust to a range of prediction distributions and achieves constant inference time regardless of dataset size. We validate our approach on both simulated and real-world data, and deploy it to measure production models at Meta. } }
Endnote
%0 Conference Paper %T Efficient Model Performance Evaluation Using a Combination of Expert and Crowd-sourced Labels %A Sam Corbett-Davies %A Viet-An Nguyen %A Udi Weinsberg %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-corbett-davies26a %I PMLR %P 4510--4518 %U https://proceedings.mlr.press/v300/corbett-davies26a.html %V 300 %X As models, particularly large language models (LLMs), are deployed on increasingly challenging tasks, correctly evaluating their performance is growing in importance and difficulty. Expert human labelers are high-quality but scarce and resource-intensive to obtain, while crowd-sourced labels are more readily accessible at scale but lower in quality. We propose Maven (Model And Voter EvaluatioN), a hierarchical Bayesian model that combines these two label sources to produce model performance estimates on binary tasks that are less biased than using crowd-sourced labels alone and have lower variance than using expert labels alone. By modeling the ranking of model scores, Maven is robust to a range of prediction distributions and achieves constant inference time regardless of dataset size. We validate our approach on both simulated and real-world data, and deploy it to measure production models at Meta.
APA
Corbett-Davies, S., Nguyen, V. & Weinsberg, U.. (2026). Efficient Model Performance Evaluation Using a Combination of Expert and Crowd-sourced Labels . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:4510-4518 Available from https://proceedings.mlr.press/v300/corbett-davies26a.html.

Related Material