论文部分内容阅读
在数据建模和分析中,有限混合体模型被广泛地使用着。然而,如何仅仅针对一组来自于某个有限混合体模型的数据选择出分量或聚类的个数则依然是一个非常困难的问题。由于分量个数是混合体模型的规模度量,其选择问题被称为有限混合体的模型选择问题。最近,针对有限混合体模型,特别是高斯混合模型,一种自动模型选择学习机制逐步发展成熟起来。这种新的机制能够在学习参数的过程中自动地完成模型选择,为数据的建模与分析提供了一种新的思路与途径。本文将对于高斯混合模型或一般有限混合体模型的自动模型选择学习算法及其典型应用进行综述与总结。首先,我们综述了基于贝叶斯阴阳机和谐学习原则的自动模型选择学习算法。然后,我们描述了另一种基于熵惩罚的自动模型选择学习算法。最后,我们给出了自动模型选择学习算法的一些典型的应用。
In data modeling and analysis, the finite mixture model is widely used. However, how to choose the quantity or the number of clusters just for a group of data from a finite mixture model remains a very difficult problem. Since the number of components is a measure of the size of a mixture model, the selection problem is called the model selection problem of a finite mixture. Recently, for the limited mixture model, especially the Gaussian mixture model, an automatic model selection learning mechanism gradually developed. This new mechanism can automatically select the model during the learning process and provides a new idea and approach for data modeling and analysis. This article reviews and summarizes the automatic model selection learning algorithms and their typical applications for Gaussian mixture models or general finite mixture models. First, we summarize the automatic model selection learning algorithm based on the Bayesian theory of harmony learning. Then, we describe another automatic model selection learning algorithm based on entropy penalty. Finally, we present some typical applications of automatic model selection learning algorithms.