Acceleration of statistical iterative algorithms for graphical models

我们利用向量epsilon算法开发了EM算法的加速方法，然后将其应用于缺失数据的可分解对数线性模型。我们还提出了基于马尔可夫基的可分解对数线性模型的MCMC。在此基础上，提出了一种基于向量epsilon算法的交替最小二乘（ALS）算法加速方法，并验证了该算法提高了交替最小二乘（ALS）算法的计算效率。

We developed the acceleration method for the EM algorithm using the vector epsilon algorithm, and then applied it to decomposable log-linear models with missing data. We also proposed MCMC using Markov bases for decomposable log-linear models. Furthermore, we developed a new acceleration method for the Alternating Least Squares(ALS) algorithm using the vector epsilon algorithm, and demonstrated that the acceleration algorithm improves the computational efficiency of the original ALS algorithm.