实现人工神经网络知识增殖能力的一种方法  

A Knowledge Increasable Method for Artificial Neural Network

作　　者：黄华[1] 罗四维[1] 李爱军[1] 刘蕴辉[1] 

机构地区：[1]北京交通大学计算机与信息技术学院,北京100044

出　　处：《计算机研究与发展》2004年第7期1062-1067,共6页Journal of Computer Research and Development

基　　金：国家自然科学基金项目 ( 69973 0 0 2 ) ;教育部博士点基金项目 ( 2 0 0 2 0 0 0 40 2 0 )

摘　　要：具有知识增殖能力的神经学习系统是人工神经网络发展的一个重要方向 ,备受研究人员的关注 传统上对神经学习系统知识的增殖或重用研究偏重于对个体网络的改造 ,根据知识积累和继承的思想 ,引入自治神经网络 (au tonomousartificialneuralnetwork ,AANN)的理念 ,以此作为构造知识可增殖神经学习系统的基础 ,利用群体网络的方法成功解决了神经学习系统的拓展和知识增殖问题 AANN和一般神经网络的区别在于其自治能力 ,采用AANN模块构造的神经学习系统 ,具有知识增殖能力 ,其可靠性、可拓展性和灵活性都得到提高 实验结果表明 。A neural learning system with the ability of knowledge increase is an important aspect of artificial neural network research,which is attracting the eyes of many machine learning researchers. Conventional research on knowledge increase/reuse of a neural learning system put an emphasis on the modification of a single network. The paper introduced the concept of autonomous artificial neural network (AANN for short) according to the idea of knowledge accumulation and succession. A neural learning system featuring knowledge increase ability is constructed with AANN modules,which is also flexible and scalable. An AANN module is different from its counterparts in that it is autonomous. With such modules as its basis,the reliability,flexibility and scalability of the system are greatly improved. Experimental results show that such a system can effectively inherit the knowledge of its components and can gain novel knowledge gradually when new modules arrive.

关 键 词：知识增殖 自治神经网络 平衡编码 机器学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]