arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2407.02119cs.LGcs.AIcs.CL
收藏

基于同策略与主动学习的高成本效益代理奖励模型构建

Cost-Effective Proxy Reward Model Construction with On-Policy and Active Learning

University of Washington(华盛顿大学);Microsoft Corporation(微软公司)

Yifang Chen, Shuohang Wang, Ziyi Yang, Hiteshi Sharma, Nikos Karampatziakis, Donghan Yu, Kevin Jamieson, Simon Shaolei Du, Yelong Shen

AI总结:

针对RLHF中人类偏好数据不足、在线算法专家查询成本高的问题,首次提出结合同策略查询与主动学习的代理奖励模型构建策略,用少量标注数据实现高效偏好标注,提升模型性能且可与其他策略融合降本。

摘要:

人类反馈强化学习(RLHF)是当前大语言模型流水线中广泛采用的方法,其发展受限于人类偏好数据的规模。传统方法依赖离线偏好数据集构建,而近期的研究方向已转向在线设置:学习器利用少量带标注种子数据和大规模未标注提示词池,通过自生成响应与高质量奖励/偏好反馈迭代构建新的偏好数据。但当前大多数在线算法仍聚焦于在给定反馈预言机的策略模型更新过程中进行偏好标注,这会产生高昂的专家查询成本。我们首次探索了成本效益型代理奖励预言机构建策略,旨在以极少的标注数据和专家查询预算完成后续偏好或奖励标注。该方法包含两项核心创新:(1)同策略查询,用于规避种子数据中的分布外(OOD)与不平衡问题;(2)主动学习,用于筛选信息量最高的数据进行偏好查询。借助这些方法,我们用最少的专家标注数据训练出一个评估模型,该模型可有效标注出9倍数量的偏好对,用于后续RLHF训练。例如,我们采用直接偏好优化(DPO)的模型仅耗费1.7K的查询成本,就在AlpacaEval2、MMLU-5shot和MMLU-0shot基准上取得了约1%以上的平均提升。我们的方法与其他基于直接专家查询的策略互不冲突,因此可与之结合以进一步降低查询成本。

英文摘要:

Reinforcement learning with human feedback (RLHF), as a widely adopted approach in current large language model pipelines, is \textit{bottlenecked by the size of human preference data}. While traditional methods rely on offline preference dataset constructions, recent approaches have shifted towards online settings, where a learner uses a small amount of labeled seed data and a large pool of unlabeled prompts to iteratively construct new preference data through self-generated responses and high-quality reward/preference feedback. However, most current online algorithms still focus on preference labeling during policy model updating with given feedback oracles, which incurs significant expert query costs. \textit{We are the first to explore cost-effective proxy reward oracles construction strategies for further labeling preferences or rewards with extremely limited labeled data and expert query budgets}. Our approach introduces two key innovations: (1) on-policy query to avoid OOD and imbalance issues in seed data, and (2) active learning to select the most informative data for preference queries. Using these methods, we train a evaluation model with minimal expert-labeled data, which then effectively labels nine times more preference pairs for further RLHF training. For instance, our model using Direct Preference Optimization (DPO) gains around over 1% average improvement on AlpacaEval2, MMLU-5shot and MMLU-0shot, with only 1.7K query cost. Our methodology is orthogonal to other direct expert query-based strategies and therefore might be integrated with them to further reduce query costs.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑