基于粗集的多层分类器的设计与实现  

Design and Implementation of a Multilayer Classifier Based on Rough Set

作　　者：许志兴 吴俊华 唐晓纹 

机构地区：[1]夏新电子股份有限公司,南京210016

出　　处：《计算机工程与应用》2006年第8期184-186,共3页Computer Engineering and Applications

摘　　要：文章提出在多层分类器中使用粗集理论来进行网络的设计,由于粗集理论有强大的数值分析能力,而多层分类器具有准确的逼近收敛能力和较高的精度,所以通过两者的结合,可以得到一种可理解性好、计算简单、收敛速度快的新型多层分类器模型。首先利用粗集理论来提取原始的领域知识,然后通过计算决策表的相对约简来产生规则,这些规则的依赖性因子被设为多层分类器的初始连接权值,这些权值在训练学习中得到改进。文章最后给出了一个决策表的实例来进一步验证了该方法的高效性和正确性。A new scheme of knowledge encoding in a multilayer classifier using rough set theoretic concepts is described in this paper.Rough Set theory has a powerful capability for qualitative analysis,while BP neural networks can approach most problems with a much satisfying accuracy.By combining those advantages of the two theories,we can construct a kind of neural network with good understandability,simple computation and exact accuracy.Firstly,Rough set theory is utilized for extracting crude domain knowledge.Then some rules are generated from a decision table by computing relative reduet.The dependency factors of these rules are encoded as the initial connection weights of the network.The network is next trained to refine its weight values.Finally,an example with satisfying results is also presented in this paper.

关 键 词：粗集 多层分类器 规则 依赖性因子 知识系统 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]