摘要
针对传统EM算法存在初始模型成分数目需要预先指定以及收敛速度随样本数目的增长而急剧减慢等问题,提出了一种快速、贪心的高斯混合模型EM算法。该算法采用贪心的策略以及对隐含参数设置适当阈值的方法,使算法能够快速收敛,从而在很少的迭代次数内获取高斯混合模型的模型成分数。该算法通过与传统EM算法、无监督EM算法和鲁棒EM算法的聚类结果进行比较,实验结果证明该算法具有很强的鲁棒性,并且能够提高算法的效率以及模型成分数的准确性。
In order to solve the disadvantages of traditional EM algorithm which initial model component parameters need to preassign and the convergence speed follows the growth of sample numbers, an rapid greedy EM algorithm for Gaussian mixture model is proposed. This algorithm adopts greedy strategy and uses appropriate implicit parameters to accelerate the speed of convergence, which can precisely get the optimal solution of the model component in a few itera-tions. In experiments, compared with traditional EM algorithm, unsupervised EM algorithm and robust EM algorithm, the algorithm is robust, in addition, it can improve the efficiency of algorithm and the accuracy of the model component parameters.
出处
《计算机工程与应用》
CSCD
北大核心
2015年第20期111-115,共5页
Computer Engineering and Applications
关键词
贪心
高斯混合模型
隐含参量
最大期望(EM)算法
greedy
Gaussian mixture model
implicit parameter
Expectation Maximization（EM）algorithm
