A Regret Minimization Framework on Preference Learning in Large Language Models
About
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 $(\textbf{RePO})$, which reframes RLHF through $\textit{regret minimization}$ rather than reward maximization. Human preferences are often shaped by $\textit{prospective}$ anticipation of outcomes and $\textit{counterfactual}$ comparisons to alternative behaviors, rather than by immediate, outcome-independent utility. $\textbf{RePO}$ 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 $\textbf{RePO}$ is an effective and human-aligned approach for training large language models.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Mathematical Reasoning | AMC 23 | Accuracy22.5 | 83 | |
| Mathematical Reasoning | GSM8K | Accuracy91.05 | 80 | |
| Preference Evaluation | AlpacaEval 2 | WR (%)60.12 | 64 | |
| Mathematical Reasoning | AMC23 | Mean Accuracy45 | 42 | |
| Human Preference Alignment | MT-Bench | -- | 20 | |
| Mathematical Reasoning | GSM8K | Pass@1 Accuracy89.49 | 19 | |
| Human Preference Alignment | Arena Hard | Win Rate (%)60.1 | 16 | |
| Mathematical Reasoning | MATH | Accuracy (Mean)63.95 | 14 | |
| Mathematical Reasoning | MATH 500 | Mean Accuracy (MATH 500)63.68 | 14 | |
| Mathematical Reasoning | Minerva | Mean Accuracy21.69 | 14 |