Multi-Objective and Mixed-Reward Reinforcement Learning via Reward-Decorrelated Policy Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Bai, Yang, Liu, Kaiyuan, Zhuang, Ziyuan, Zhou, Jiahong, Weng, Rongxiang, Chen, Xin, Wang, Jingang, Cai, Xunliang
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913123633463296
author Bai, Yang
Liu, Kaiyuan
Zhuang, Ziyuan
Zhou, Jiahong
Weng, Rongxiang
Chen, Xin
Wang, Jingang
Cai, Xunliang
author_facet Bai, Yang
Liu, Kaiyuan
Zhuang, Ziyuan
Zhou, Jiahong
Weng, Rongxiang
Chen, Xin
Wang, Jingang
Cai, Xunliang
contents Complex reinforcement learning environments frequently employ multi-task and mixed-reward formulations. In these settings, heterogeneous reward distributions and correlated reward dimensions often destabilize the construction of scalar advantages. To address these challenges, we propose Reward-Decorrelated Policy Optimization (RDPO), a reward-processing method designed to explicitly target both failure modes. RDPO first utilizes Magnitude-Aware Quantile normalization to stabilize prompt-level advantage allocation across binary, fractional, and continuous rewards. It then applies Mahalanobis whitening within each active reward subspace to mitigate correlation redundancy prior to aggregation. When applied during the post-training of LongCat-Flash, RDPO enhances instruction following, writing quality, and robustness to hard prompts while remaining broadly competitive on reasoning and coding evaluations.
format Preprint
id arxiv_https___arxiv_org_abs_2605_13641
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Multi-Objective and Mixed-Reward Reinforcement Learning via Reward-Decorrelated Policy Optimization
Bai, Yang
Liu, Kaiyuan
Zhuang, Ziyuan
Zhou, Jiahong
Weng, Rongxiang
Chen, Xin
Wang, Jingang
Cai, Xunliang
Machine Learning
Computation and Language
Complex reinforcement learning environments frequently employ multi-task and mixed-reward formulations. In these settings, heterogeneous reward distributions and correlated reward dimensions often destabilize the construction of scalar advantages. To address these challenges, we propose Reward-Decorrelated Policy Optimization (RDPO), a reward-processing method designed to explicitly target both failure modes. RDPO first utilizes Magnitude-Aware Quantile normalization to stabilize prompt-level advantage allocation across binary, fractional, and continuous rewards. It then applies Mahalanobis whitening within each active reward subspace to mitigate correlation redundancy prior to aggregation. When applied during the post-training of LongCat-Flash, RDPO enhances instruction following, writing quality, and robustness to hard prompts while remaining broadly competitive on reasoning and coding evaluations.
title Multi-Objective and Mixed-Reward Reinforcement Learning via Reward-Decorrelated Policy Optimization
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2605.13641