报告题目：Efficient Split Likelihood Method for Community Detection of Large-scale Networks
报 告 人：刘秉辉 副教授 东北师范大学
报告时间：2019年12月10日上午9：00-10：00
报告地点：数学楼第二报告厅
报告摘要：
To recover community labels under the stochastic block model (SBM), we propose a split likelihood (SL) framework, which aims at providing a rapidly converging algorithm with advantages in terms of both the accuracy of community detection and computational efficiency. Under such framework, we create an alternative inference function, the split likelihood, to avoid handling the problem of the intractability of the inference of the likelihood of the original observation, by splitting variables of the original SBM into two independent split bodies with identical distribution. Then, we create some effective computing strategies to maximize the split likelihood. Based on them, we propose the efficient SL algorithm and establish its computational and statistical properties. We demonstrate the superiority of the proposed methods via some numerical results as well as a real data analysis.
报告人简介：
刘秉辉，东北师范大学，副教授，博士生导师，统计系主任；主要研究方向为应用统计、机器学习、网络数据分析；在Artificial Intelligence、Journal of Machine Learning Research、The Annals of Applied Statistics、Statistics in Medicine、Computational Statistics & Data Analysis等期刊发表多篇学术论文；主持国家自然科学基金青年项目一项、面上项目一项，主持中央高校基本科研业务费青年拔尖人才项目一项，主持吉林省科技厅重点实验室专项课题一项；与中国联通合作开发大数据产品一项、主持大数据培训项目一项；主持开发长春市市长公开电话数据挖掘项目热点分析子模块。
