核聚类快速后向传播算法  

Fast kernel clustering back propagation algorithm

作　　者：孙全玲[1] 王立新[1] 

机构地区：[1]安徽建筑工业学院电子与信息工程学院计算机工程系,合肥230601

出　　处：《计算机工程与应用》2013年第10期118-120,162,共4页Computer Engineering and Applications

基　　金：安徽省教育厅自然科学基金(No.KJ2011B041)

摘　　要：后向传播神经网络算法是一种经典的分类算法,但是通常该算法训练时间较长。针对这种不足,提出了一种基于核聚类的快速后向传播算法。利用核聚类将原始样本划分为多个簇,对每一个簇计算簇中心样本,利用所有的簇中心样本作为新训练集进行神经网络学习。在UCI标准数据集和说话人识别数据集上的仿真实验,充分说明了算法较传统后向传播算法具有明显的速度优势。The back propagation algorithm is a classic classification algorithm, but it is usually with a long training time. For this deficiency, this paper presents a fast back propagation algorithm based on kernel clustering. The algorithm uses kernel clustering to divide the original samples into multiple clusters, then computes the sample’s center of each cluster, and uses all the center samples as the new training set and trains a neural network classifier. Simulation experiments on UCI standard data set and speaker recognition data set show that the proposed algorithm has obvious advantages compared with the traditional back propagation algorithm.

关 键 词：后向传播 神经网络 核聚类 说话人识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]