基于Kullback-Leibler与PCA的概率密度比值估计  

Ratio Estimation of Probability Density Based on Kullback-Leibler and PCA

作　　者：兰远东[1,2] 邓辉舫[1] 

机构地区：[1]华南理工大学计算机科学与工程学院,广东广州510000 [2]惠州学院计算机科学系,广东惠州516007

出　　处：《计算机技术与发展》2012年第6期107-110,共4页Computer Technology and Development

基　　金：国家自然科学基金(61170193)

摘　　要：为了更好地解决在机器学习和数据挖掘等领域中经常遇到的两个概率密度函数的比值估计问题,文中提出了一种新的概率密度比值估计算法。该算法基于Kullback-Leibler距离,综合混合高斯模型和主成分分析的概率密度比值估计方法,使用混合概率主成分分析为两个概率密度比值函数建模。在概率密度比值估计的过程中,不是分别估计比值函数的分子和分母,而是对整个比值函数进行混合组成建模。算法避免了分别对分子分母的概率密度估计,降低了估计的误差。实验表明该算法能够获得较好的估计结果。In order to solve the estimation problems for the ratio of two probability density functions, which is often encountered in ma- chine learning and data mining area,propose a new estimation algorithm for the ratio of the probability density. The algorithm is based on Kullback-Leibler distance, integrated Ganssian mixture model and principal component analysis of the estimation methods, use mixed probabilistic principal component analysis for the modeling. In the estimation process for the ratio of the probability density, not separately estimate the numerator and denominator of the ratio function, but modeling the function in the same time. In this way, the algorithm can reduce the estimated error. Experiments show that the algorithm can obtain better result.

关 键 词：概率密度 机器学习 主成分分析 样本空间 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]