基于鲁棒高斯混合模型的加速EM算法研究

发布时间：2018-03-29 04:06

本文选题：EM算法　切入点：鲁棒　出处：《计算机应用研究》2017年04期

【摘要】：针对传统鲁棒高斯混合模型EM算法存在模型成分参数难以精确获取最优解以及收敛速度随样本数量的增加而快速降低等问题,提出了一种基于鲁棒高斯混合模型的加速EM算法。该算法采用隐含参量信息熵原理对高斯模型分量个数进行挑选,以及使用Aitken加速方法减少算法的迭代次数,当接近最优解时,EM步长的变化极为缓慢,这时使用Broyden对称秩1校正公式进行校正,使算法快速收敛,从而能够在很少的迭代次数内精确获取高斯混合模型的模型成分数。该算法通过与传统鲁棒EM算法和无监督的EM算法的聚类结果进行比较,实验证明该算法对初始值的设定并不敏感(成分数c无须预先设定),并且能够降低算法运算时间,提高聚类模型成分数(类簇)的正确率。
[Abstract]:In view of the traditional EM algorithm with robust Gao Si mixed model, it is difficult to obtain the optimal solution accurately by the component parameters of the model and the convergence rate decreases rapidly with the increase of the sample number. In this paper, an accelerated EM algorithm based on robust Gao Si mixed model is proposed, which adopts the principle of implicit parameter information entropy to select the number of components of Gao Si model, and uses Aitken acceleration method to reduce the number of iterations of the algorithm. When approaching the optimal solution, the change of EM step size is very slow, then the Broyden symmetric rank 1 correction formula is used to correct the algorithm, which makes the algorithm converge quickly. Thus the model of Gao Si hybrid model can be accurately obtained in a few iterations. The algorithm is compared with the clustering results of the traditional robust EM algorithm and the unsupervised EM algorithm. The experimental results show that the algorithm is not sensitive to the initial value setting (the fractional c does not need to be set in advance), and it can reduce the computation time and improve the accuracy of clustering model into fractions (clusters).
【作者单位】：辽宁工程技术大学电子与信息工程学院;
【基金】：国家自然科学基金资助项目(61402212) 辽宁省高等学校杰出青年学者成长计划资助项目(LJQ2015045) 辽宁省自然科学基金资助项目(2015020098) 辽宁省教育厅一般资助项目(L2013131)
【分类号】：TP301.6

【相似文献】