带混合属性的神经网络规则提取方法

人工神经网络应用中最大的弊端是缺乏可理解性,而对结果的解释是任何一个完善的智能系统必备的基本特征.从神经网络中提取规则被公认为是解决该问题最有效的手段之一.因此,所提取规则的可理解程度成为衡量规则提取算法质量的重要指标.目前该领域的研究主要集中在分类规则的提取上.对于分类问题,待测模式的属性的取值可能是离散的,也可能是连续的.现有的算法针对全连续或者全离散的问题已取得较好的效果.但对既包含连续属性也包含离散属性的问题,已有算法未取得理想的结果.本文针对带混合属性的分类问题,提出了一种规则提取算法,在提取规则的可理解性上同时照顾了连续属性和离散属性.

Method for the Identification of Rules in the Nerve Net with Complex Property

In the application of manual nerve net,the lacking of understandability is the most serious shortcoming,so the ability to explain the result is the essential characteristic for all the capable intelligent systems.The identification of rules from the nerve net is recognized as one of the most efficient methods to solve such problem.Therefore,the understandability of the identified rules becomes the important index to evaluate the arithmetic quality of the identified rules.Presently,the main study of this fie...