面向数据流的人工免疫网络增量学习技术  

Incremental learning approach of artificial immune network in data streams

作　　者：岳训[1,2] 迟忠先[1] 莫宏伟[3] 郝艳友[1] 

机构地区：[1]大连理工大学计算机科学工程系,辽宁大连116024 [2]山东农业大学信息科学与工程学院,山东泰安271018 [3]哈尔滨工程大学自动化学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2006年第B07期247-251,共5页Journal of Harbin Engineering University

基　　金：国家自然科学基金资助项目（60305007）

摘　　要：针对大规模、动态、时变的数据流环境，给出了一种基于内影像记忆网络的人工免疫网络增量学习算法ICainet，当数据增加时，通过计算新添加数据与记忆网络的匹配程度给出不同的增量处理策略，从而使新算法具有动态跟踪不断递增的大规模数据流的增量学习能力．由于新算法已对数据流先进行一定程度的数据压缩并能“记忆以前数据”，因而占用内存空间少、运算代价很低．实验结果表明，在损失较小程度学习质量的前提下，该方法在运行时间上有明显的优势，可满足不同的应用环境需求．A new Incremental Immune network ICainet in data steams environment is proposed. It uses an ＂internal image＂ memory network to reflect the dynamic changes of data streams. By the similarity between new data and the ＂internal image＂ memory network, new data are added to the data collection by two main strategies to address the problem of incremental learning. It has the ability of allowing dynamic tracking of ever-increasing large scale information without having to perform complete reclustering. The results of experimental evaluation show that the new approach provides significantly faster data summarization than complete re-clustering does. Based on losing little quality of clustering, it has significant superiority on running time. So it can meet the needs of solving the on-line large scale dynamic data clustering in different applications of data streams.

关 键 词：人工免疫网络模型 增量学习 数据流 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]