---
id: 20261029-T0-17
title: "CONCUR：多任务多奖励强化学习中的一致性辅助奖励"
title_en: "CONCUR: Concordant Auxiliary Rewards for Multi-Task, Multi-Reward RL"
url: https://ai.daily.yangsir.net/daily/20261029-T0-17
issue_date: 2026-10-29
publish_date: 2026-10-28T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2610.36437
---

# CONCUR：多任务多奖励强化学习中的一致性辅助奖励

arXiv新论文研究LLM在多任务强化学习中的奖励设计问题。当一个任务的主奖励失败时，长度等辅助奖励无法超过主奖励的上限，导致辅助信号在失败样本上失效。论文提出CONCUR方法，用一致性辅助奖励来处理多任务、多奖励场景下的信号冲突。论文为预印本，具体基准与提升幅度需查阅原文。

## English Version

**CONCUR: Concordant Auxiliary Rewards for Multi-Task, Multi-Reward RL**

A new arXiv paper addresses reward design when an LLM is trained with reinforcement learning across multiple tasks. A task's primary reward can gate auxiliary rewards such as length: on a failing response, the auxiliary cannot exceed the primary, weakening the learning signal. CONCUR proposes concordant auxiliary rewards to handle these conflicts in multi-task, multi-reward settings. The paper is a preprint; benchmarks and gains are detailed in the full text.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2610.36437)

**详情页**：https://ai.daily.yangsir.net/daily/20261029-T0-17

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*