cs.AIJun 8, 2026

A Regret Minimization Framework on Preference Learning in Large Language Models

Authors: Suhwan KimTaehyun ChoGeon-Hyeong KimYu Jin KimYoungsoo JangMoontae LeeJungwoo Lee

Abstract

Reinforcement learning with verifiable rewards (RLVR) has enabled progress on reasoning-intensive tasks by relying on task-specific verifiers that provide automated correctness signals. However, many realistic language tasks are difficult to equip with reliable verifiers, motivating a growing reliance on reinforcement learning from human feedback (RLHF). In this setting, we argue that a closer examination of how human feedback should be interpreted is essential. We introduce Regret-based Preference Optimization (RePO)(\textbf{RePO}), which reframes RLHF through regret minimization\textit{regret minimization} rather than reward maximization. Human preferences are often shaped by prospective\textit{prospective} anticipation of outcomes and counterfactual\textit{counterfactual} comparisons to alternative behaviors, rather than by immediate, outcome-independent utility. RePO\textbf{RePO} captures this structure by modeling preferences as behavior-conditioned assessments of relative suboptimality. Experiments on mathematical reasoning benchmarks and human preference datasets demonstrate consistent performance gains, indicating that RePO\textbf{RePO} is an effective and human-aligned approach for training large language models.

Explore similar work

CardsList
  1. General Preference Reinforcement Learning

    May 18, 2026Muhammad Umer, Muhammad Ahmed Mohsin, Ahsan Bilal +5Deep Reinforcement Learning