CMStatistics 2017: Start Registration
View Submission - CMStatistics
Title: A stochastic EM algorithm for maintenance modeling with random improvement factors Authors:  Min Gong - City University of Hong Kong (Hong Kong) [presenting]
Abstract: Maintenance policy changes the lifetime behavior of a system by introducing one or more factors into the underlying model, named after improvement factors. Due to many uncontrollable reasons, different maintenance actions should associate with different improvement factors, so these factors should be treated as random. However, random factors lead to great difficulty in parameter estimation. EM algorithm is a standard approach to estimate parameter for model with hidden variables. We compute the expected value of the complete log-likelihood function with respect to the distribution of the hidden variable in the E step, and then maximize it in the M step. The maximum likelihood estimate of parameters can be found by iterating these two steps. However, traditional EM algorithm has the pitfall that the rate of convergence can be painfully slow. Moreover, when the number of simulated samples is large, the maximization step is intractable. To improve the convergence behavior of the Monte Carlo EM algorithm, we here employ the idea of stochastic EM algorithm. Because the expectation in E step is an intractable integral, for certain high-dimensional problems, MCMC simulation is the only known technique capable of providing a solution within a reasonable time. We herein develop some samplers based on the Metropolis-Hastings algorithm, check its feasibility and make comparison.