一种半监督机器学习的EM算法改进方法

EM(Expectation Maximization)算法是含有隐变量(latent variable)的概率参数模型最大似然估计、极大后验概率估计最有效的算法,但很容易进入局部最优现象,对此提出基于半监督机器学习机制的EM算法.本文方法是在最大似然函数中加入惩罚最小二乘因子,同时引入非负约束作为先验信息,结合半监督机器学习方法,将EM算法改进转化为最小化求解问题,再采用最大似然方法求解EM模型,有效估计了混合矩阵和高斯混合模型参数,实现EM算法的改进.仿真结果表明,该方法能够很好地解决了EM算法容易局部最优化问题.

Improved EM Algorithm of Semi-supervised Machine Learning

EM algorithm is the most effective algorithm for maximum likelihood estimation and maximum posteriori probability estimation of probability parameter model with latent variable,but it is easy to enter the local optimum phenomenon.An EM algorithm based on semi-supervised machine learning is proposed.The least squares penalty term is added to the maximum likelihood function,non-negative constraints are introduced as a priori information,and combined with semi-suervised machine learning method,the improved EM algorithm is transformed into a minimization problem.The maximum likelihood algorithm is used to solve the EM model,effectively estimating the parameters of the mixed matrix and the Gaussian mixed model,and the improvement of the EM algorithm is realized.The simulation results show that this method can solve the problem of EM algorithm easily entering local optimum well.