Strategies for High Performance Graph-Based Reasoning

高性能基于图的推理策略

• 批准号: 0412854
• 负责人: Rina Dechter
• 金额: --
• 依托单位: University of California-Irvine
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2004
• 资助国家: 美国
• 起止时间: 2004-09-01 至 2008-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0412854&HistoricalAwards=false
• 关键词: Strategies; Performance; Graph; Based; Reasoning

This project seeks to develop adaptive techniques for high performance graph-based reasoning systems that allow users to control the tradeoffs between computational resources and solution quality. The main thrust of this project is to introduce adaptability and scalability in algorithms for constraint optimization, probabilistic inference, and decision making under uncertainty. The project is structured into subprojects that study: (1) iterative belief propagation for graphical models; (2) hybrids of stochastic local search and inference; (3) search guided by partition-based heuristics; and (4) mixed probabilistic and deterministic (constraint) networks. These subprojects are tied together by the PI's ongoing research on the unifying framework of "parameterized bounded inference" that combines the two paradigms of search and structure-based inference. Endowing graph-based algorithms with increased adaptability and scalability is important not only to progress in AI and computer science but also to application in many domains. An additional goal of this project is to package the developed algorithms in one software reasoning and evaluation shell (REES) to allow uniform empirical evaluation and to facilitate dissemination of the project's results by researchers, educators and application builders.

该项目旨在为基于图形的推理系统开发自适应技术，这些技术使用户可以控制计算资源和解决方案质量之间的权衡。该项目的主要目的是在算法中引入适应性和可伸缩性，以进行约束优化，概率推断和不确定性下的决策。该项目被构成研究：（1）图形模型的迭代信念传播； （2）随机局部搜索和推理的杂种； （3）以基于分区的启发式方法为指导的搜索； （4）混合概率和确定性（约束）网络。这些子项目由PI对“参数化有限推理”的统一框架进行的持续研究结合在一起，该研究结合了搜索和基于结构的推断的两个范式。赋予具有提高适应性和可伸缩性的基于图的算法不仅对AI和计算机科学的进展以及在许多领域的应用都很重要。该项目的另一个目标是在一个软件推理和评估外壳（REES）中包装开发的算法，以允许统一的经验评估，并促进研究人员，教育工作者和应用程序建筑商对项目结果的传播。