Integration of Deductive Learning and Inductive Learning by Extended EBL

本研究的目的是建立一种在不完全领域理论下获取有效和有用的宏观规则的方法。得到以下结果：1)提出了一种扩展的EBL。作为可操作性准则，引入了使用率最大化和回溯次数最小化两个准则。最小EBG对于递增地寻找运算泛化是有用的。利用增量式最小EBG生成器实现了一个学习系统。2)通过引入最大覆盖的概念，提出了一种解决不一致问题的方法。通过对常见解释结构的泛化层次进行自上而下的搜索，可以找到包含所有正例而排除所有反例的最佳宏观规则集。3)一种解决棘手问题的方法聚焦于一类具有连续可分解子目标的问题解决，即从示例中获取问题解决宏表的增广EBL学习者。通过对八大谜题的应用，证明了该学习者的有效性。

The objective of this research was to establish a methodology for acquiring valid and useful macro rules under the imperfect domain theory.The following results were obtained1)proposition of an extended EBLThe augmented EBL is a framework for knowledge refinement based on generalization from prural examples. As operationality criteria, the maximization of the usage and the minimization of backtracking number were introduced. The least EBG is useful to find operational generalizations incrementally. A Learning system with an incremental least EBG generator has been realized.2)an approach to the inconsistent problemA method for solving the inconsistent problem has been proposed by introducing a concept of maximal covering. By top down search over generalization hierarchy of common explanation structures, the best set of macro rules which includes all positive examples and excludes all negative ones can be found.3)an approach to the intractable problemFocusing an a class of problem solving that have serially desomposable subgoals, an augmented EBL learner that acquires a problem solving macrotable from examples. The usefulness of the learner has been shown by applying to the eight puzzle problem.