基于知识规则的构造性优先排序神经网络算法  

Learning algorithm model of constructive PONN based on knowledge rules

作　　者：谷秧波[1] 武妍[1] 朱世交[1] 王守觉[2] 

机构地区：[1]同济大学计算机科学与工程系,上海200092 [2]同济大学半导体与信息技术研究所

出　　处：《计算机应用》2008年第7期1645-1648,共4页journal of Computer Applications

基　　金：国家自然科学基金资助项目(60475019)

摘　　要：从高维空间样本点覆盖的角度,讨论了基于知识规则的构造性优先排序神经网络(PONN)算法的原理,提出了网络构造过程的一般算法以及基于随机取样规则和重心点规则的两个实例算法。实例算法对螺旋线识别和语种识别进行了仿真。实验结果证明了算法的有效性。语种识别实验结果也表明基于重心规则的PONN算法在一定条件下优于SVM。From the perspective of covering points in high-dimensional space, we discussed the theory of constrictive PONN based on knowledge rules. PONN＇s general constructive algorithm and its two special methods based on random sampling and barycenter rules respectively were proposed in this paper. Simulations on spirals recognition and voice language identification were conducted using the special methods. Experimental results illustrate the good performance of PONN instances. Besides, voice ：language identification results show that PONN method based on barycenter rule outperforms SVM under certain circumstances.

关 键 词：优先度排序神经网络 知识规则 覆盖 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]