基础模型(FMs)已经革新了计算的许多领域,包括自动规划与调度(APS)。例如,最近一项研究发现它们对规划问题很有用:计划生成、语言翻译、模型构建、多智能体规划、交互式规划、启发式优化、工具集成以及类脑规划。除 APS 之外,还有许多看似相关的任务,涉及生成一系列动作,并对这些动作实现预期目标的可执行性具有不同程度的保证;我们将这些任务统称为类规划(PL)任务,例如业务流程、程序、工作流和指南,研究人员已经考虑在其中使用 FMs。然而,以往工作主要关注预训练的现成 FMs,并选择性地对其进行微调。本文讨论了从零开始构建一个面向 PL 任务的综合性 FM 的必要性,并探讨了其设计考量。我们认为,这样的 FM 将为 PL 问题求解开辟新的高效途径,正如 LLM 正在为 APS 所创造的那样。
英文摘要:
Foundation Models (FMs) have revolutionized many areas of computing, including Automated Planning and Scheduling (APS). For example, a recent study found them useful for planning problems: plan generation, language translation, model construction, multi-agent planning, interactive planning, heuristics optimization, tool integration, and brain-inspired planning. Besides APS, there are many seemingly related tasks involving the generation of a series of actions with varying guarantees of their executability to achieve intended goals, which we collectively call planning-like (PL) tasks like business processes, programs, workflows, and guidelines, where researchers have considered using FMs. However, previous works have primarily focused on pre-trained, off-the-shelf FMs and optionally fine-tuned them. This paper discusses the need for a comprehensive FM for PL tasks from scratch and explores its design considerations. We argue that such an FM will open new and efficient avenues for PL problem-solving, just like LLMs are creating for APS.
Combining Offline Models and Online Monte-Carlo Tree Search for Planning from Scratch
Xiamen University(厦门大学)
随机和部分可观测环境中的规划问题是人工智能的一个核心问题。解决这类问题的一个常用技术是首先构建一个精确的模型。虽然最近已经提出了一些方法来学习模型不确定性下的最佳行为,先验知识的环境仍然需要保证所提出的算法的性能。利用预测状态表示(Predictive State Representations~(PSR))方法的状态表示和模型预测的优点,本文提出了一种从头开始规划的方法,该方法首先离线学习PSR模型,然后结合在线蒙特卡罗树搜索进行模型不确定性规划。通过与模型不确定性规划方法的比较,证明了所提出方法的有效性及其收敛性。我们所提出的方法的有效性和可扩展性也进行了测试的RockSample问题,这是不可行的最先进的BA-POMDP为基础的方法。
AppForge: From Assistant to Independent Developer -- Are GPTs Ready for Software Development?
Key Lab of HCST (PKU), MOE(HCST关键实验室(PKU));School of Computer Science, Peking University, China(北京大学计算机科学学院);Columbia University, USA(哥伦比亚大学);Beijing Jiaotong University, China(北京交通大学);University of Texas at Dallas, USA(德克萨斯大学达拉斯分校);Hong Kong University of Science and Technology, Hong Kong SAR, China(香港科技大学);Fudan University Institute of Systems for Advanced Computing, Shanghai, China(复旦大学先进计算系统研究所);Simon Fraser University, Canada(西蒙弗雷泽大学);Shanghai Institute of Systems for Open Computing, Shanghai, China(上海开放计算系统研究所)
Deep Learning for Generalised Planning with Background Knowledge
LAAS-CNRS(法国国家科学研究中心实验室分析与建筑系统实验室);University of Toulouse(图卢兹大学);The Australian National University(澳大利亚国立大学);Czech Technical University in Prague(布拉格捷克理工大学)