arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2404.05902cs.CLcs.AI
收藏

WILBUR:面向鲁棒且准确的 Web 智能体的自适应上下文学习

WILBUR: Adaptive In-Context Learning for Robust and Accurate Web Agents

University of California Berkeley(加州大学伯克利分校);Bardeen, Inc.(巴登公司)

Michael Lutz, Arth Bohra, Manvel Saroyan, Artem Harutyunyan, Giovanni Campagna

AI总结:

针对 Web 智能体跨网站泛化与准确性难题,本文提出 Wilbur,通过可微排序、指令合成、智能回溯和自动课程数据训练,在 WebVoyager 上取得最先进结果。

摘要:

在 Web 智能体研究领域,同时实现泛化性与准确性仍是一个具有挑战性的问题。由于网站结构差异很大,现有方法往往会失败。此外,现有的微调和上下文学习技术无法在多个网站之间泛化。我们提出 Wilbur,该方法使用可微排序模型和新颖的指令合成技术,利用来自先前运行的任务演示,以最优方式填充黑盒大语言模型的提示。为了最大化端到端成功率,我们还提出了一种智能回溯机制,能够从错误中学习并恢复。最后,我们表明排序模型可以在来自生成式自动课程的数据上训练;该课程从 LLM 采样具有代表性的目标、运行智能体并自动评估,无需人工标注。Wilbur 在 WebVoyager 基准上取得了最先进结果,总体上比纯文本模型高 8%,在某些网站上最高高出 36%。在同一基准上,尽管 Wilbur 只接收文本输入,其表现仍与一个强大的多模态模型相差不超过 5%;进一步分析表明,大量失败源于操作 Web 时的工程挑战。

英文摘要:

In the realm of web agent research, achieving both generalization and accuracy remains a challenging problem. Due to high variance in website structure, existing approaches often fail. Moreover, existing fine-tuning and in-context learning techniques fail to generalize across multiple websites. We introduce Wilbur, an approach that uses a differentiable ranking model and a novel instruction synthesis technique to optimally populate a black-box large language model's prompt with task demonstrations from previous runs. To maximize end-to-end success rates, we also propose an intelligent backtracking mechanism that learns and recovers from its mistakes. Finally, we show that our ranking model can be trained on data from a generative auto-curriculum which samples representative goals from an LLM, runs the agent, and automatically evaluates it, with no manual annotation. Wilbur achieves state-of-the-art results on the WebVoyager benchmark, beating text-only models by 8% overall, and up to 36% on certain websites. On the same benchmark, Wilbur is within 5% of a strong multi-modal model despite only receiving textual inputs, and further analysis reveals a substantial number of failures are due to engineering challenges of operating the web.

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。

↑