Learning Search-Control Heuristics for Logic Programs: Applications to Speedup Learning and Language Acquisition

逻辑程序的学习搜索控制启发式：加速学习和语言习得的应用

• 批准号: 9310819
• 负责人: Raymond Mooney
• 金额: $ 19万
• 依托单位: University of Texas at Austin
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1994
• 资助国家: 美国
• 起止时间: 1994-03-01 至 1997-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9310819&HistoricalAwards=false
• 关键词: Learning; Search; Control; Heuristics; Logic

9310819 Mooney A general goal of machine learning is automating the construction of efficient knowledge-based systems. Current methods tend to focus on either the acquisition of the basic domain knowledge (inductive learning) or the acquisition of search-control knowledge (speedup learning). However, these two tasks are not always cleanly separated. This research investigates the general framework of learning search-control heuristics for logic programs, which can improve both accuracy and efficiency. Logic programming provides a very general and well-understood representational and computational platform on which to build. Recent methods in inductive logic programming can be used to automatically learn search-control heuristics from a set of training problems. Efficiency may be improved by learning control rules that eliminate backtracking during the search for a proof. Accuracy may be improved by learning control rules that eliminate search paths that produce incorrect results.

小行星9310819 机器学习的一般目标是自动构建高效的基于知识的系统。 目前的方法往往集中在基本领域知识的获取（归纳学习）或搜索控制知识的获取（加速学习）。 然而，这两项任务并不总是完全分开的。 本研究探讨逻辑程式学习搜寻-控制演算法的一般架构，以提高演算法的准确度与效率。 逻辑编程提供了一个非常通用和易于理解的表示和计算平台。 归纳逻辑编程中的最新方法可以用来从一组训练问题中自动学习搜索控制算法。 可以通过学习在搜索证明期间消除回溯的控制规则来提高效率。 可以通过学习消除产生不正确结果的搜索路径的控制规则来提高准确性。