拟合HErD的EM算法.PPTVIP

四、拟合HErD的EM算法 五、实验结果 五、实验结果 2016.12.9 龚红仿 A Novel Approach for Phase-Type Fitting with the EM Algorithm IEEE TRANS. ON DEPENDABLE AND SECURE COMPUTING, VOL. 3, NO. 3, 2006 Department of Telecommunications, Budapest University of Technology and Economy, Hungary. 一、研究内容 A novel approach was presented that fits a restricted class of phase-type distributions, namely, mixtures of Erlang distributions, to trace data. For the parameter fitting, an algorithm of the EM was developed. To illustrate the effectiveness of the proposed fitting algorithm, the paper presented comparative results for this approach and two other methods using six benchmark traces and two real traffic traces as well as quantitative results from queueing analysis. 二、研究背景 1. 在通信模型中，构造分析上的跟踪模型，以获得跟踪数据最重要的统计特性。 2.为了分析性能和可靠性模型，跟踪数据不得不用PH分布近似拟合。根据采样数据或者其它已知分布估计PH分布参数称为PH拟合。 3.EM算法是普遍的解决最大似然估计方法。 二、研究背景 问题：1. 用EM算法拟合一般的PH分布变得非常昂贵，而且拟合分布十分依赖于所给的初值。 2.用EM算法拟合四阶以上的PH分布似乎并不合适，并且比较小的变异系数的跟踪数据不能由低阶PH分布表示。 解决方法：用于拟合的PH分布类必须通过初始化矩阵中的非零元而限制在基本的EM算法中；或者开发EM算法适应于特定类型的PH分布。 二、研究背景 Fit the parameters of a hyperexponential distributionan using an EM algorithm [8]. However, hyperexponential distributions cannot adequately capture general distributions with increasing and decreasing densities or with a coefficient of variation less than one. 二、研究背景 Apart from EM, other optimization algorithms can also be applied. However, the optimization problem for general PH distributions is too complex to yield satisfactory results if the number of phases is larger than two or three. The fitting problem becomes much easier if acyclic instead of general PH distributions are used. 二、研究背景 For the acyclic PH distributions, the resulting optimization is still complex and contains local optima and saddle points. To overcome the problem of convergence to a local optimum, the fitting algorithm is usually started with several initial settings and the best fitting is chosen. 二、研究背景 In this paper, an EM algorithm for the fitting of hyper-E