Automatic Learning of Admissible Heuristic Evaluation Functions

可接受的启发式评估函数的自动学习

• 批准号: 9619447
• 负责人: Richard Korf
• 金额: $ 25.08万
• 依托单位: University of California-Los Angeles
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1997
• 资助国家: 美国
• 起止时间: 1997-05-15 至 2002-04-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9619447&HistoricalAwards=false
• 关键词: Automatic; Learning; Admissible; Heuristic; Evaluation

The objective of this project is to develop a robust set of algorithms for automatically learning admissible heuristics - that is, functions that identify optimal solutions to combinatorial problems and give a lower bound on the actual cost of a solution. The more accurate the heuristic function, the more efficiently an optimal solution can be found, or if an optimal solution is not required the more quickly a lower-cost solution can found. This research is developing ways to learn these heuristics in a domain-independent manner by applying a technique that has proved useful in automatically learning more accurate admissible heuristic functions that take more time to compute, but make up for it in the time they save for the algorithm using the heuristic. This work will provide better solutions of combinatorial problems of artificial intelligence which will be increasingly cost-effective as computers get faster and larger problems are presented for solution.

这个项目的目标是开发一套健壮的算法，用于自动学习可接受的启发式算法，即识别组合问题的最佳解决方案并给出解决方案实际成本的下界的函数。启发式函数越精确，找到最优解的效率就越高，或者如果不需要最优解，找到成本较低的解的速度就越快。这项研究正在开发一种方法，通过应用一种技术，以一种独立于领域的方式来学习这些启发式，这种技术已被证明在自动学习更准确的可接受的启发式函数方面很有用，这些函数需要更多的时间来计算，但可以在使用启发式算法节省的时间中弥补它。这项工作将为人工智能组合问题提供更好的解决方案，随着计算机的速度越来越快，问题越来越大，人工智能组合问题的成本效益将越来越高。