学术报告十七:邹长亮—Large-Scale Datastreams Surveillance via Pattern-Oriented-Sampling

发布时间:2020-09-21

报告题目:Large-Scale Datastreams Surveillance via Pattern-Oriented-Sampling

时  间 2020年9月20日(星期日)上午10:00

地  点翡翠湖校区翡翠科教楼B座1710

报告人邹长亮教授

工作单位: 南开大学统计与数据科学学院

举办单位bb电子糖果派对

报告人简介邹长亮,南开大学统计与数据科学学院教授。2008年于南开大学获博士学位,随后留校任教。主要从事统计学及其与数据科学领域的交叉研究和实际应用。研究兴趣包括:高维数据统计推断、大规模数据流分析、变点和异常点检测等,在Ann.Stat.、Biometrika、J.Am.Stat.Asso.、Math. Program.、Technometrics、IISE Tran.等统计学和工业工程领域期刊上发表论文几十篇,主持国家自然科学基金委重大项目课题、国家杰青项目、国家优青项目等。

报告摘要Monitoring large-scale datastreams with limited resources has become increasingly important for real-time detection of abnormal activities in many applications. Despite the availability of large datasets, the challenges associated with designing an efficient change-detection when clustering or spatial pattern exists are not yet well addressed. In this talk, I will introduce a design-adaptive testing procedure when only a limited number of streaming observations can be accessed at each time. We derive an optimal sampling strategy, the pattern-oriented-sampling, with which the proposed test possesses asymptotically and locally best power under alternatives. Then, a sequential change-detection procedure is proposed by integrating this test with generalized likelihood ratio approach. Benefiting from dynamically estimating the optimal sampling design, the proposed procedure can improve the sensitivity in detecting clustered changes compared with existing procedures. Its advantages are demonstrated in numerical simulations and a real data example. Ignoring the neighboring information of spatially structured data will tend to diminish the detection effectiveness of traditional detection procedures.