arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2601.17879cs.CLcs.AIcs.IR
收藏

自管理:用于长篇深度研究的并行代理循环

Self-Manager: Parallel Agent Loop for Long-form Deep Research

University of Chinese Academy of Sciences(中国科学院大学);ModelBest Inc.(ModelBest公司);University of Queensland(昆士兰大学);University of California Merced(加州大学默塞德分校)

Yilong Xu, Zhi Zheng, Xiang Long, Yujun Cai, Yiwei Wang

AI总结:

Self-Manager通过并行代理循环提升长篇深度研究的效率和灵活性,实现异步执行和上下文隔离,优于现有单代理方法。

摘要:

长篇深度研究需要在更长的范围内进行多方面的调查以获得全面报告。在处理此类复杂任务时,现有代理在子任务级别管理上下文以克服线性上下文积累和信息丢失。然而,它们仍然遵循单一上下文窗口和顺序执行范式,这导致相互干扰和阻塞行为,限制了可扩展性和适应性。为了解决这个问题,本文介绍了Self-Manager,一种并行代理循环,能够实现异步和并发执行。主线程可以创建多个子线程,每个线程都有自己的隔离上下文,并通过线程控制块进行迭代管理,从而实现更专注和灵活的并行代理执行。为了评估其有效性,我们将在DeepResearch Bench上对Self-Manager进行基准测试,结果显示它在所有指标上都优于现有的单代理循环基线。此外,我们进行了广泛的分析实验,以证明Self-Manager设计选择的必要性,以及其在上下文容量、效率和泛化方面的优势。

英文摘要:

Long-form deep research requires multi-faceted investigations over extended horizons to get a comprehensive report. When handling such complex tasks, existing agents manage context at the subtask level to overcome linear context accumulation and information loss. However, they still adhere to a single context window and sequential execution paradigm, which results in mutual interference and blocking behavior, restricting scalability and adaptability. To address this issue, this paper introduces Self-Manager, a parallel agent loop that enables asynchronous and concurrent execution. The main thread can create multiple subthreads, each with its own isolated context, and manage them iteratively through Thread Control Blocks, allowing for more focused and flexible parallel agent execution. To assess its effectiveness, we benchmark Self-Manager on DeepResearch Bench, where it consistently outperforms existing single-agent loop baselines across all metrics. Furthermore, we conduct extensive analytical experiments to demonstrate the necessity of Self-Manager's design choices, as well as its advantages in contextual capacity, efficiency, and generalization.

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。

↑