arXiv:2501. 07437v3 Announce Type: replace-cross Abstract: Most statistical models for pairwise comparisons, including the Bradley-Terry (BT) and Thurstone models and many extensions, make a relatively strong assumption of stochastic transitivity.
arXiv:2608. 02081v1 Announce Type: new Abstract: In this paper, we study prediction problems for paired comparison data, for example, predicting the win probability between two unmatched players and ranking all the players according to the order of their strengths by using win probability data between two matched players.
The paper introduces a low‑rank framework for ranking large language models (LLMs) on task‑specific benchmarks using sparse pairwise comparisons. By modeling the task‑by‑model ability matrix as low rank, the method shares information across related tasks while preserving task‑specific differences, and it provides uncertainty‑aware ranking through debiased estimators and simultaneous confidence sets. Experiments on synthetic data and the Chatbot Arena benchmark demonstrate improved sample efficiency and tighter, better‑calibrated ranking certificates, especially in the sparse comparison regime typical of real LLM evaluations.
arXiv:2608. 10045v1 Announce Type: cross Abstract: The problem of learning from pairwise comparisons has been widely studied across many domains such as recommendation systems, social choice, and more recently, fine-tuning large language models.
arXiv:2608.25200v2 Announce Type: replace-cross Abstract: We study learning a mixture of $k$ Plackett-Luce models from multi-way ranking responses from annotators that may represent heterogeneous und...
arXiv:2604. 17805v2 Announce Type: replace-cross Abstract: Pairwise ranking systems based on Maximum Likelihood Estimation (MLE), such as the Bradley-Terry model, are widely used to aggregate preferences from pairwise comparisons.
The paper treats large language model (LLM) evaluation as a tensor completion problem, modeling noisy, sparse, and non‑uniform pairwise human judgments through a low‑rank latent score tensor under Bradley‑Terry‑Luce‑type models. It derives the efficient influence function and semiparametric efficiency bound for smooth functionals of the true tensor, and proposes a one‑step debiased estimator with asymptotic normality. A key innovation is a score‑whitening technique that equalizes local Fisher information, overcoming anisotropy in the information operator and enabling stable inference at optimal sample‑complexity.
Among chess opening positions that a strong engine judges essentially equal (Stockfish 18 evaluation within 10 centipawns of zero, depth-stable) and that humans actually reach on Lichess (October 2025; 1,661 positions, 16. 1M occurrences), human results are not balanced.
arXiv:2509. 23102v4 Announce Type: replace Abstract: Reinforcement learning from human feedback (RLHF) has emerged as the standard paradigm for aligning large language models with human preferences.
arXiv:2607. 08979v1 Announce Type: new Abstract: We study the active learning problem of fixed-confidence top-$k$ identification from noisy pairwise comparisons.
arXiv:2602. 03061v2 Announce Type: replace-cross Abstract: Evaluating mathematical reasoning in LLMs is constrained by limited benchmark sizes and inherent model stochasticity, yielding high-variance accuracy estimates and unstable rankings across platforms.
arXiv:2607. 25655v1 Announce Type: new Abstract: Among chess opening positions that a strong engine judges essentially equal (Stockfish 18 evaluation within 10 centipawns of zero, depth-stable) and that humans actually reach on Lichess (October 2025; 1,661 positions, 16.
arXiv:2510. 15824v2 Announce Type: replace-cross Abstract: This article considers an online version of conformal inference, called adaptive conformal inference [ACI] and introduced by Gibbs and Cand\`es (2021): prediction sets are issued sequentially, after observing features and before the outcomes are revealed.