Calibrated Preference Learning: The Case of Label Ranking

Santo M. A. R. Thies, Viktor Bengs, Timo Kaufmann, Sebastian Josef Vollmer, Eyke Hüllermeier
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:120504-120543, 2026.

Abstract

Calibration, the alignment of predicted probabilities with true outcome frequencies, is essential for reliable decision-making. While extensively studied for classification and regression, calibration has not been formally addressed for probabilistic label ranking, where the goal is to predict a distribution over orderings of a label set. Naively treating rankings as classes ignores their structure and fails to capture important modalities such as pairwise and top-k predictions. We formalize calibration for label ranking and develop a hierarchy of notions covering full rankings, sub-rankings, and top-k rankings. We prove that full-rank calibration implies the others but not conversely, and sub-ranking and top-k calibration are incomparable. Empirically, we find popular label ranking models are often poorly calibrated, with substantial differences between sub-ranking and top-k metrics. Applying our framework to RLHF reward models, we find that calibration correlates strongly but not perfectly with benchmark accuracy, suggesting it captures a meaningful quality dimension beyond top-1 accuracy. These findings motivate future work on understanding the downstream effects of miscalibration and developing methods to correct it.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-thies26a, title = {Calibrated Preference Learning: The Case of Label Ranking}, author = {Thies, Santo M. A. R. and Bengs, Viktor and Kaufmann, Timo and Vollmer, Sebastian Josef and H\"{u}llermeier, Eyke}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {120504--120543}, 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/thies26a/thies26a.pdf}, url = {https://proceedings.mlr.press/v306/thies26a.html}, abstract = {Calibration, the alignment of predicted probabilities with true outcome frequencies, is essential for reliable decision-making. While extensively studied for classification and regression, calibration has not been formally addressed for probabilistic label ranking, where the goal is to predict a distribution over orderings of a label set. Naively treating rankings as classes ignores their structure and fails to capture important modalities such as pairwise and top-k predictions. We formalize calibration for label ranking and develop a hierarchy of notions covering full rankings, sub-rankings, and top-k rankings. We prove that full-rank calibration implies the others but not conversely, and sub-ranking and top-k calibration are incomparable. Empirically, we find popular label ranking models are often poorly calibrated, with substantial differences between sub-ranking and top-k metrics. Applying our framework to RLHF reward models, we find that calibration correlates strongly but not perfectly with benchmark accuracy, suggesting it captures a meaningful quality dimension beyond top-1 accuracy. These findings motivate future work on understanding the downstream effects of miscalibration and developing methods to correct it.} }
Endnote
%0 Conference Paper %T Calibrated Preference Learning: The Case of Label Ranking %A Santo M. A. R. Thies %A Viktor Bengs %A Timo Kaufmann %A Sebastian Josef Vollmer %A Eyke Hüllermeier %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-thies26a %I PMLR %P 120504--120543 %U https://proceedings.mlr.press/v306/thies26a.html %V 306 %X Calibration, the alignment of predicted probabilities with true outcome frequencies, is essential for reliable decision-making. While extensively studied for classification and regression, calibration has not been formally addressed for probabilistic label ranking, where the goal is to predict a distribution over orderings of a label set. Naively treating rankings as classes ignores their structure and fails to capture important modalities such as pairwise and top-k predictions. We formalize calibration for label ranking and develop a hierarchy of notions covering full rankings, sub-rankings, and top-k rankings. We prove that full-rank calibration implies the others but not conversely, and sub-ranking and top-k calibration are incomparable. Empirically, we find popular label ranking models are often poorly calibrated, with substantial differences between sub-ranking and top-k metrics. Applying our framework to RLHF reward models, we find that calibration correlates strongly but not perfectly with benchmark accuracy, suggesting it captures a meaningful quality dimension beyond top-1 accuracy. These findings motivate future work on understanding the downstream effects of miscalibration and developing methods to correct it.
APA
Thies, S.M.A.R., Bengs, V., Kaufmann, T., Vollmer, S.J. & Hüllermeier, E.. (2026). Calibrated Preference Learning: The Case of Label Ranking. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:120504-120543 Available from https://proceedings.mlr.press/v306/thies26a.html.

Related Material