← 研踪 PaperTrailarXiv:2611.25446 · v1
PAIR-MED: Policy-reward Alternation for Iterative Refinement in Multilingual Medical Reasoning
首次提交 2026-11-19 11:25
版本更新 2026-11-19 11:25
雷达观测 2026-11-20 10:50
摘要
Large language models demonstrate strong medical question-answering capabilities, yet reasoning performance remains uneven across languages. Process reward models (PRMs) provide step-level feedback for improving reasoning, but a fixed PRM may become increasingly misaligned with the trajectory distribution of an evolving policy during reinforcement learning, thereby reducing the reliability of its supervision. We propose a policy--reward alternating training framework that jointly adapts a medical PRM and optimizes a multilingual reasoning policy. Following supervised initialization, the current policy generates multilingual trajectories to update the PRM, which then provides process-level rewards for the next policy update. Medical step supervision maintains the PRM's medical foundation, while process, outcome, and language-related rewards guide policy learning. We develop multilingual medical PRMs (PAIR-PRM) and reasoning models (PAIR-MED) at 8B and 9B scales across 12 languages. On the medical subsets of Global-MMLU and MMLU-ProX, our models improve average accuracy by 3.01--7.47% over their corresponding Qwen backbones. On CUREMED-BENCH, PAIR-MED-9B improves the joint language consistency and logical accuracy metric by 12.88% and 7.45% over baselines for in-domain and out-of-domain languages. Further analyses examine PRM adaptation, multilingual reasoning quality, and the contributions of individual training components.
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-19 11:25 · 最新观测版本
PAIR-MED: Policy-reward Alternation for Iterative Refinement in Multilingual Medical Reasoning
Large language models demonstrate strong medical question-answering capabilities, yet reasoning performance remains uneven across languages. Process reward models (PRMs) provide step-level feedback for improving reasoning, but a fixed PRM may become increasingly misaligned with the trajectory distribution of an evolving policy during reinforcement learning, thereby reducing the reliability of its supervision. We propose a policy--reward alternating training framework that jointly adapts a medical PRM and optimizes a multilingual reasoning policy. Following supervised initialization, the current policy generates multilingual trajectories to update the PRM, which then provides process-level rewards for the next policy update. Medical step supervision maintains the PRM's medical foundation, while process, outcome, and language-related rewards guide policy learning. We develop multilingual medical PRMs (PAIR-PRM) and reasoning models (PAIR-MED) at 8B and 9B scales across 12 languages. On the medical subsets of Global-MMLU and MMLU-ProX, our models improve average accuracy by 3.01--7.47% over their corresponding Qwen backbones. On CUREMED-BENCH, PAIR-MED-9B improves the joint language consistency and logical accuracy metric by 12.88% and 7.45% over baselines for in-domain and out-of-domain languages. Further analyses examine PRM adaptation, multilingual reasoning quality, and the contributions of individual training components.
查看此版本解读 →查看此版本原文 ↗