统计与管理学院2017年学术报告第76期
【主 题】 A Diagnostic Procedure for High-Dimensional Data Streams Via Missed Discovery Rate Control
【报告人】 濮晓龙 教授
华东师范大学
【时 间】 2017年12月19日（星期二）10:00-11:00
【地 点】 上海财经大学统计与管理学院大楼1208会议室
【摘 要】In monitoring complex systems, apart from quick detection of abnormal changes of system performance, accurate fault diagnosis of responsible variables has become critical in many applications that involve high-dimensional data streams. Conventional statistical process control (SPC) diagnostic methods are often computationally expensive. More importantly, as the assumption that only one or a few variables are out-of-control (OC) is invalid for high-dimensional data streams, the fact that they cannot control the missed discovery rate (MDR) will be a major drawback. In this paper, we frame fault isolation as a multiple-testing problem to provide a diagnosis framework by controlling a novel weighted MDR at some level. The use of weights provides an effective strategy to incorporate information on the shift size in large-scale inference. Given the oracle optimality and the data-driven optimality asymptotically, the diagnostic result can be obtained easily and quickly. Simulation results and a real-data analysis from a semiconductor manufacturing process are presented to demonstrate the effectiveness of our method.
