学术报告

学术报告

您当前所在位置: 首页 > 学术报告 > 正文
报告时间 2023年4月23日下午14:30—16:00 报告地点 南校区会议中心121会议室
报告人 孙毅

报告题目:An efficient convex hull finding algorithm for graphical models

报告人:孙毅 教授 新疆大学

报告时间:2023年4月23日下午14:30—16:00

报告地点:南校区会议中心121会议室

报告人简介:孙毅,新疆大学数学学院教授,硕师生导师;2016年6月博士毕业于南开大学应用数学专业;2016年10月进入东北师范大学博士后流动站,从事统计学研究。现主要研究领域为概率图模型。在Applied Mathematics and Computation,Ramanujan Journal, Discrete Mathematics and Theoretical Computer Science, Discrete Applied Mathematics等杂志发表30余篇科研论文;主持国家自然科学基金项目2项,博士后面上项目1项,自治区项目3项。

报告摘要:A convex hull containing the variables of interest in a graph is the minimal convex subgraph containing them, which has been turned out to be a simple and intuitive characterization of the collapsibility in graphical models. In this talk, we propose the Minimal Separator Absorption Algorithm (MSA), a low-order polynomial time complexity algorithm that efficiently identifies the unique convex hull containing a set of variables. The MSA algorithm is based on the relationship between convex subgraphs and minimal separators. By absorbing proximal minimal separators, we propose an accelerated algorithm called the Proximal Minimal Separator Absorption Algorithm (PMSA), to find the unique convex hull containing variables of interest on which the statistical inference problem can be solved. Experimental results demonstrate that our approach significantly outperforms state-of-the-art algorithms. Finally, we apply PMSA to two examples to demonstrate the practicality of our proposed algorithm for structural dimension reduction.

主办单位:伟德国际BETVlCTOR

上一篇:Introduction to shrinking gradient Ricci solitons

下一篇:面向层次结构数据的学习初探

关闭