The Sign Estimator: Preference Modeling for LLM Alignment under Heterogeneity

Ali Aouad, Aymane El Gadarri, Vivek Farias
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:3103-3133, 2026.

Abstract

Traditional large language model (LLM) alignment methods are based on Reinforcement Learning From Human Feedback (RLHF), which learns a single reward model (implicitly or explicitly) from pairwise comparison data. This approach implicitly assumes homogeneous preferences across human labelers—an assumption that is violated in practice. As a result, the learned reward model in RLHF is generally misspecified: Prior work shows that it is inconsistent with the population-average utility, incurring large distortion, and that recovering this utilitarian objective is provably impossible in the worst case. In this work, we show that the average utility is recoverable under a mild assumption. Our method, the Sign Estimator, simply replaces the standard cross-entropy loss function with a notion of binary classification loss and yields a reward model that is ordinally consistent with the population-average utility. We further establish a fast finite-sample convergence rate of $O({n^{-{1}/{3}}})$ , which provides, to our knowledge, the first consistent estimator for heterogeneous preferences that does not suffer from the curse of dimensionality.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-aouad26a, title = {The Sign Estimator: Preference Modeling for {LLM} Alignment under Heterogeneity}, author = {Aouad, Ali and Gadarri, Aymane El and Farias, Vivek}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {3103--3133}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/aouad26a/aouad26a.pdf}, url = {https://proceedings.mlr.press/v306/aouad26a.html}, abstract = {Traditional large language model (LLM) alignment methods are based on Reinforcement Learning From Human Feedback (RLHF), which learns a single reward model (implicitly or explicitly) from pairwise comparison data. This approach implicitly assumes homogeneous preferences across human labelers—an assumption that is violated in practice. As a result, the learned reward model in RLHF is generally misspecified: Prior work shows that it is inconsistent with the population-average utility, incurring large distortion, and that recovering this utilitarian objective is provably impossible in the worst case. In this work, we show that the average utility is recoverable under a mild assumption. Our method, the Sign Estimator, simply replaces the standard cross-entropy loss function with a notion of binary classification loss and yields a reward model that is ordinally consistent with the population-average utility. We further establish a fast finite-sample convergence rate of $O({n^{-{1}/{3}}})$ , which provides, to our knowledge, the first consistent estimator for heterogeneous preferences that does not suffer from the curse of dimensionality.} }
Endnote
%0 Conference Paper %T The Sign Estimator: Preference Modeling for LLM Alignment under Heterogeneity %A Ali Aouad %A Aymane El Gadarri %A Vivek Farias %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-aouad26a %I PMLR %P 3103--3133 %U https://proceedings.mlr.press/v306/aouad26a.html %V 306 %X Traditional large language model (LLM) alignment methods are based on Reinforcement Learning From Human Feedback (RLHF), which learns a single reward model (implicitly or explicitly) from pairwise comparison data. This approach implicitly assumes homogeneous preferences across human labelers—an assumption that is violated in practice. As a result, the learned reward model in RLHF is generally misspecified: Prior work shows that it is inconsistent with the population-average utility, incurring large distortion, and that recovering this utilitarian objective is provably impossible in the worst case. In this work, we show that the average utility is recoverable under a mild assumption. Our method, the Sign Estimator, simply replaces the standard cross-entropy loss function with a notion of binary classification loss and yields a reward model that is ordinally consistent with the population-average utility. We further establish a fast finite-sample convergence rate of $O({n^{-{1}/{3}}})$ , which provides, to our knowledge, the first consistent estimator for heterogeneous preferences that does not suffer from the curse of dimensionality.
APA
Aouad, A., Gadarri, A.E. & Farias, V.. (2026). The Sign Estimator: Preference Modeling for LLM Alignment under Heterogeneity. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:3103-3133 Available from https://proceedings.mlr.press/v306/aouad26a.html.

Related Material