Journal of Software
1000-9825
1996
7
4
233
237
article
解释学习结果的优化排序
THE RESULT ORGANIZATION IN EXPLANATION─BASED LEARNING
解释学习可以直接应用于知识库的推理加速．经解释学习学到的一组规则，其参加匹配的先后次序直接影响着学习的效用．本文提出并证明了在以规则的成功频率与本身匹配费用的比值为序自大至小排列时，可以使这些规则的整体匹配费用最小．文中给出学习结果的优化排序算法，对于训练例的任意给定次序，可求得最小费用排列．
In this paper,a simple expression E,i.e.,a term or an atom,is viewed as a set,the"so-called"generated set of E,of the Herbrand universe U.or Herbrand base BLof the language L.The authors give a criteria to determine whether an expression coversanother one,and present the notion of decomposition expressions and the decomposition algorithm on which a necessary and sufficient condition for the covering relation between one and several other expressions is based.Then,they show how to partition a union of the sets generated by the simple expressions according to the partition formula in the set theory.Finally,they generalize generis these above results into the conjunctions through defining a transformation which converts a conjunction into a simple expression.This paper provides the theoretic foundation of the procedural semantics of authors'annotated logic program.
解释学习; 效用分析; 知识库加速; 规则排序;
Explanation-based learning; utility analysis; knowledge base speedup; ordering of rules.;
郝继刚,石纯一
Hao Jigang and Shi Chunyi
jos/article/abstract/19960406