一种概率映射网络的EM训练算法

文中提出一种概率映射网络(PMN)的EM(Expectation Maximization)训练算法。PMN为一个四层前馈网。它构成一个贝叶斯分类器，实现多类分类的贝叶斯判别，把输入的样本模式经网络变换为输出的分类判决，其网络节点对应于贝叶斯后验概率公式的各个变量。 此PMN用高斯核函数作为密度函数，网络参数训练由EM算法实现，其学习方式为类间的监督学习和类内的非监督学习。最后的实验表明此网络及其学习算法在分类应用中的有效性。

AN EFFICIENT EM TRAINING ALGORITHM FOR PROBABILITY MAPPING NETWORKS

An Expectation-Maximization(EM) training algorithm for estimating the parameters of a special Probability Mapping Network (PMN) structure which forms a multicatolog Bayes classifier is proposed in this paper. The structure of PMN is a four-layer Feedforward Neural Networks(FNN), where the Gaussian probability density function is realized as an internal node. In this way, the EM algorithm is extended to deal with supervised learning of a multicatolog of the neural network Gaussian classifier. The computational efficiency and the numerical stability of the training algorithm benefit from the well- established EM framework. The effectiveness of the proposed network architecture and its EM training algorithm are assessed by conducting two experiments.