题目：Dynamic Modeling and Online Monitoring of Tensor Data Streams with Application to Passenger Flow Surveillance
汇报人：吴纯杰
会议时间：2022年12月15日（周四） 18:00
地点：综合楼644会议室
交流平台：腾讯会议 664-118-615
报告人简介：

吴纯杰，上海财经大学统计与管理学院教授、副院长、博士生导师，中国青年统计学家协会常务理事，中国现场统计研究会可靠性工程分会常务理事和中国商业统计学会理事等。研究领域为应用统计和政府统计，在统计质量管理、市场满意度测评、客户流失分析和金融建模等方面开展研究工作。在国内外权威期刊Journal of the American Statistical Association，IISE Transactions和《中国科学：数学》，《经济学季刊》等发表论文40余篇，主持国家自然科学基金项目2项和国家统计局重大项目1项；上海市精品课程、课程思政示范课程、教学团队和国家级一流本科课程《数理统计》负责人，国家级一流统计学专业建设点负责人，主持完成上海市重点课程1项和承担上海市教学项目3项，获得上海市教学、科研奖3项；指导学生作品获得“挑战杯”全国一等奖等国家级和省部级奖项20余项。

摘要：

Dynamic tensor data streams are becoming prevalent in various application domains such as traffic networks, smart manufacturing, etc. It is crucial to monitor such tensor data streams to promptly detect abnormal activities and system failures. Existing tensor monitoring methods rely heavily on the assumption that the tensor coefficients exhibit a low-rank structure or are inapplicable to general-order tensors. In this article, prompted by the surveillance of subway passenger flow, we propose a unified framework for the dynamic modeling and online monitoring of tensor data streams. Based on the tensor normal distribution, we first derive a tensor model selection procedure, through which the selected tensor structure strikes a balance between model complexity and estimation accuracy. Then we propose an online estimation procedure to dynamically estimate the tensor process's parameters, based on which sequential change-detection procedures are proposed using the generalized likelihood ratio test. By benefiting from modeling the complex correlation structure of different tensor modes dynamically, our procedures can improve the sensitivity with which various types of changes are detected by comparison with existing methods. The efficacy of our approach is illustrated by extensive simulations and analysis of real passenger flow surveillance data.