报告题目:tSEM: Tensor Structural Equation Modeling for Latent Cross-Modal Association
报告人:史建清 教授
报告时间:2026年11月3日(周二)10:00-11:00
报告地点:藕舫楼724室
主持人:曹春正 教授
报告人简介:

史建清,英国皇家统计学会会士, 科技部十四五重点项目首席科学家,北师香港浸会大学统计与数据科学讲座教授,研究生院执行院长。曾任南方科技大学统计与数据科学系和深圳国家应用数学中心教授,理学院生物医学统计中心主任,英国国家艾伦图灵研究院图灵研究员,剑桥大学牛顿学院访问研究员,英国纽卡斯尔大学统计学教授,纽卡斯尔大学云计算和大数据研究中心副主任。主要研究方向包括健康大数据分析,函数型数据分析,生物医学统计,缺失数据分析,Meta-analysis等。在国际学术刊物上发表高水平学术论文130多篇,包括统计和医学顶级期刊 JRSSB, JASA, Biometrika, Nature Medicine 和British Medical Journal。现任J. of Computational and Graphical Statistics和Statistical Methods & Applications等期刊副主编;曾任英国皇家统计协会《应用统计》(JRSSC)等国际期刊副主编,并担任该刊Discussion Paper的特邀副主编。获IEEE康复游戏和健康国际年会最佳论文奖、美国统计协会非参数统计分会年度最佳论文奖。在Chapman & Hall 出版专著:Gaussian Process Regression Analysis for Functional Data.
报告简介:
In many scientific applications, a central goal of multimodal analysis is to understand how latent patterns behind different data modalities are related. When one modality is tensor-valued and another is vector-valued, a key challenge is to preserve the multiway structure of the tensor modality while explicitly characterizing its association with latent factors from the vector-valued modality. We propose a tensor structural equation modeling (tSEM) framework for tensor-vector multimodal data. The proposed tSEM model embeds a CP measurement equation into a structural equation modeling framework, so that tensor-derived latent scores can be linked to vector-modality latent factors through a structural coefficient matrix. We further introduce a sparse extension, tSESM, by adding a sparsity penalty to the structural matrix, encouraging a more interpretable representation of cross-modal associations. The model parameters are estimated jointly through a composite criterion that combines tensor reconstruction, cross-modal structural fitting, and covariance-matching regularization. As an application, we analyze EEG tensor data and questionnaire measurements from insomnia patients and controls, illustrating that the proposed framework can extract physiologically interpretable EEG components and provide a structural map linking them to questionnaire-derived symptom factors.
欢迎广大师生踊跃参加!
数学与统计学院
江苏省应用数学(南京信息工程大学)中心
江苏省系统建模与数据分析国际合作联合实验室
江苏省统计科学研究基地
2026年10月30日