报告人：郭建华 (东北师范大学)
报告题目：Profile-marginal likelihood methods for community detection of large-scale networks
报告摘要：In recent years, network science has been developing rapidly in many fields. One very important task of network analysis is that of detecting communities, which are key to understanding the structure of complex networks. Detecting communities is often considered as a challenging problem, due to high computational complexity. Among a large number of existing approaches to solving this problem, statistical ones have grown increasingly popular. The stochastic block models are generally viewed as the most studied statistical models for modeling community structure and performing community detection. Most of the existing algorithms aiming to fit the stochastic block models have difficulties in dealing with large-scale networks, with the exception of only a few methods including a very popular one, named the pseudo-likelihood method, which performs community detection by maximizing a substitute of the observational likelihood. Though it has achieved important progresses, it still has obvious insufficiencies: it is not theoretically convergent, and not very suitable for small-scale networks. On this ground, we propose a novel statistical model-based method, named the profile-marginal likelihood method, which adopts a hybrid framework of the profile and the marginal likelihoods. As suggested by a large number of theoretical and numerical results, as well as a real data analysis, the proposed method retains all the advantages of the pseudo-likelihood method, and can overcome both of the insufficiencies. In particular, it has significant advantages on both community detection accuracy and computation efficiency, especially for large-scale sparse networks. Interestingly, the proposed method shows outstanding extension ability in dealing with networks with degree heterogeneity within communities and bipartite networks.
报告人简介：郭建华，东北师范大学副校长，教授，博士生导师。国务院学位委员会学科评议组统计学科召集人，国家杰出青年科学基金获得者，国务院政府特殊津贴获得者，国家社会科学基金学科规划评议组成员，国家自然科学基金会评专家，著名期刊《JASA》、《统计研究》等的编委。
报告时间：2018年6月26日（星期二）下午4:30- 5:30.
报告地点：科技楼南楼702
