报告题目:Birth-death MCMC graphical model selection

报告摘要: Regression and graphical model are two important statistical tools in data science and statistical genetics. While under high-dimensional setting, which is very common in this big data era, the model selection is a serious problem. Lasso and some other varieties are classical model selection methods in Frequentist inference, but these methods usually suffer two problems: the estimation of likelihood function and the choice of penalty parameters. On the other hands, in Bayesian model selection, the computation of Bayes factor, and the model search in high dimensional problems are also difficult to handle. In this paper, we will modify the Birth-death MCMC(BDMCMC) method and apply it to regression and graphical model problems. With the BDMCM method, we can quickly get samples from the approximated posterior distribution of model given data, then Bayesian model averaging is used to select the important covariates or interactions.

报 告 人：王南伟博士 多伦多大学

报告时间：2019年5月6日（星期一）下午16：00~16：50

报告地点：数学科学学院38号楼报告厅

主办单位：数学科学学院

欢迎广大师生参加！