Holder： Zhengjun Zhang（University of Chinese Academy of Sciences）
Time：2025-02-27 15:10-17:00
Location：Room 217, Guanghua Building 2
Abstract:
Understanding causal relationships among variables is crucial in economic, biological, medical, climate, and many other applied research fields. Conventional methods often struggle with asymmetric causality and high-dimensional data. To address these challenges from a machine learning perspective, this talk introduces MMSPE-HMAC—a Minimum Mean Squared Prediction Error (MMSPE) Hamiltonian-clustering Modernized Asymmetric Causality (HMAC) method. MMSPE-HMAC integrates Generalized Measures of Correlation (GMC) into deep clustering with a RadViz-style representation, utilizing an optimal Hamiltonian cycle to map clusters, similarities, and outliers. This enables clear visualization of causal relationships, offering a significantly different representation from existing approaches. Under the MMSPE principle, we theoretically justify that GMC leads to an optimal causative method.
Compared to other causal inference techniques, MMSPE-HMAC requires the fewest structural and statistical assumptions. It is widely applicable, easily implementable, and empirically interpretable. Extensive experiments across synthetic, engineering, machine learning, economic, and financial datasets demonstrate MMSPE-HMAC's superior performance over existing methods. Notably, MMSPE-HMAC reveals that USD/CNY exchange rate changes drive movements in USD/EUR, USD/GBP, and USD/JPY, while also identifying annual block timing effects in macroeconomic indicators. Furthermore, MMSPE-HMAC uncovers indirect causal effects in MNIST and fashion designs—patterns that are difficult to detect using other causal methods. Joint work with Tianyi Huang and Shenghui Cheng.
About the Speaker:
张正军教授现为中国科学院大学经济与管理学院长聘教授和统计与数据科学系系主任，中国科学院预测科学研究中心副主任，原美国威斯康辛大学统计系终身教授和系副主任，威斯康辛大学生物医学信息系兼职教授，国际数理统计协会执行委员和财务总监（July 2016 -- July 2022），国际数理统计协会会士，美国统计协会会士。现担任JASA，JBES， Statistica Sinica， JDS， EJS、STaRF等国际期刊副主编。主要研究方向包括统计理论和方法、计量经济学、金融计量学、计算医学与实践、 极端气候等等。在国际顶级期刊：统计（AoS，JASA，JRSSB）、计量（JoE, EE）、金融（JBES, JBF）、医学（AFM， Vaccines，npj Precision Oncology）、气象 (ATM) 等发表论文上百篇。代表性工作和首创性思想和作品包括: 新极值理论、绝对和相对同步有效性（AbRelaTEs）、双边截断极值惩罚变量选择机器学习模型（TWT-LR-ETP）、商相关系数（QCC、TQCC）、非对称广义相关系数（GMC）、滞后尾部相依系数(lambda_k)、最大线性回归模型(MaxLR)、最大逻辑回归模型(Max-logistic)、EGB2期权定价公式、盯市在险价值（MMVaR）、条件极值Frechet自回归（AcF）, 虚拟标准数字货币（VSTC），新冠基因组学、癌症基因组学的几何空间（DARPA： Mathematical Challenge Fifteen: The Geometry of Genome Space），等等。
Your participation is warmly welcomed!
欢迎扫码关注北大统计科学中心公众号，了解更多讲座信息!
