珠海数学杰出学者讲座第38讲-Aggregated Projection Method: A New Approach for Group Factor Model

讲座题目：Aggregated Projection Method: A New Approach for Group Factor Model
讲座时间 Datetime: 2024年10月18日，周五，16:30开始
地点 Venue: 海琴二号457
报告人Speaker: 王学钦 教授
单位 Affiliation: 中国科学技术大学
主持人 Host：赵育林 教授
报告摘要 Abstract:
Abstract: In addressing the challenge of distinguishing global from group-specific factors within the group factor model, especially when faced with strongly correlated local factors, this paper proposes a novel objective function. This function is designed to optimize the identification process by maximizing the sum of correlations between global and group factors, utilizing eigen-decomposition of the aggregated projection matrix that these factors form. Our method is not only computationally efficient but also robust against the distortions caused by closely correlated local factors. We establish a solid theoretical foundation for our approach, demonstrating the consistency of our global factor number estimation, and the consistency and asymptotic distributions of the estimated global/local factors and their loadings. Through extensive simulation studies, we prove our method's superior performance over existing approaches across various scenarios. Applied to the analysis of the growth rates of house prices in the United States, our method effectively identifies a primary global factor significantly influencing national house price trends, along with several local factors that highlight unique state-level impacts. This application not only showcases the practical utility of our approach but also deepens our understanding of economic dynamics through factor analysis.
报告人简介：
王学钦，中国科学技术大学讲席教授、教育部“CJ学者奖励计划”特聘教授。曾获得国家自然科学基金优秀青年基金支持，入选教育部新世纪优秀人才支持计划，国际统计学会推选会员 (ISI Elected Member），教育部高等学校统计学类专业教学指导委员会委员，并担任统计学国际期刊JASA、SII等的编委。曾获高等学校科学研究优秀成果奖自然科学二等奖（排名第一）。主要从事人工智能的统计学理论、方法与计算，特别是最优子集选择问题，非欧（几里得）数据的统计推断理论基础、与基于机理和数据融合的统计建模和推断等方向的研究。在包括统计学/机器学习等领域的知名期刊JASA、AOS、JRSSB、PNAS、IJOC、JMLR、Nature Genetics等发表科研论文100多篇论文。发布了Ball，abess等多个开源R或Python软件包，总下载量已超过120万次。