104學年度第二學期專題討論統計學術演講公告
統計碩士學位學程將於104年4月12日(星期二)舉辦統計學術演講，邀請到成功大學統計系陳瑞彬教授蒞臨演講，
講題：Bayesian Sparse Support Union Recovery
講員：陳瑞彬 教授 成功大學統計系
時間：2016年4月12日星期二 下午 3：30 ~ 5：30
地點：臺灣大學博雅教學館408教室
摘要：
In this paper, instead of the single response model, the multiple response regression model is considered. In addition to the sparsity assumption, it is future assumed that these multiple responses share the similar active variables which are selected from a candidate set. The goal is to recover the union of the sets of active variables (supports) for each individual response. Here Bayesian variable selection methods are adopted by introducing two nested sets of binary indicator variables. The first set of indicator variables is used to denote whether variables are active for any of the response vectors. The second set of indicator variables is associated with both predictor variables and responses, and each indicator is used to indicate whether the corresponding variable is active for a particular response vector. Thus to recovery the union of the supports can be solved based on the posterior samples of these two sets of indicator variables. Several MCMC algorithms are proposed for posterior sampling and the median probability criterion is used for posterior inference. The simulation results show that the proposed Bayesian methods outperform the related Lasso type methods. In addition to the simulation studies, the performances of the proposed Bayesian methods are also illustrated by real examples in image representation.
歡迎對於此議題有興趣的師生同仁參加。現場坐位有限，敬請及早報到入場。
聯絡人：張仲凱
聯絡方式：E-mail：ntustat@ntu.edu.tw
TEL：(02) 3366-1481 FAX：(02) 3366-1482