连续学习分类系统研究  

Research on Continuous Learning Classifier System

作　　者：顾大千[1] 高阳[1] 

机构地区：[1]南京大学计算机软件新技术国家重点实验室,南京210093

出　　处：《复旦学报（自然科学版）》2004年第5期765-768,共4页Journal of Fudan University：Natural Science

基　　金：国家自然科学基金资助项目(60103012);国家重点研究发展规划973项目(2002CB312002);江苏省创新人才项目(BK2003409)

摘　　要：学习分类系统(LCS)是一个动态感应环境的模拟认知系统,它利用环境反馈评估种群中的分类规则并通过遗传算法对种群进行进化.当环境输入包含连续属性时,经典LCS无法遍历整个状态空间.提出一种新的基于神经网络的连续学习分类系统,并通过实验验证了这种连续学习分类系统能够较准确地进行连续属性离散化,从而提高系统分类精度.Learning classifier system (LCS) is an adaptive learning system.LCS evaluate classifiers by the feedback from environment with the help of reinforcement learning, and use genetic algorithm in an evolutionary process. LCS can't scan all the state space when condition attributes contain continuous values, so continuous learning classifier system is one of the major aspects in LCS research. It gives a brief introduction on LCS and several measures used in continuous attribute question, advances a new continuous LCS, in which continuous attributes are dispersed by neural network first, at the end is the experiments and conclusion.

关 键 词：连续属性离散化 分类系统 分类规则 遍历 状态空间 神经网络 遗传算法 实验验证 LCS 反馈 

分 类 号：N031[自然科学总论—科学技术哲学] O172[理学—数学]