COLLEGE OF ARTS AND SCIENCES Department of Mathematics and Statistics
(click here for colloquia)

Mathematics Colloquium: Nonparametric Cross-Dimensional Inference for High-Dimensional Dependent Data


4:10pm Webster Hall 11

Xiongzhi Chen

Modern data in genomics, genetics, neuroscience, image processing, text mining, and finance often contain measurements of a very large number of features that are subject to systematic internal interdependencies or external perturbations. As such, they are high-dimensional and highly dependent. This poses great challenges for scalable, accurate and reproducible inference for such data. Existing methods may be unscalable, incapable to deal with complicated dependence, or suboptimal. To mitigate these issues, we propose a “Nonparametric Cross-Dimensional Inference” methodology that leverages the blessing of dimensionality and latent variable structures for simultaneous inference for high-dimensional dependent data. Our method is accurate, fast, flexible and scalable. It complements the expectation-maximization algorithm and variational Bayes methods, and can be optimal. Its excellent performances are supported by theory, simulation studies, and applications to population genetics, RNA-seq gene expression study in yeast, and recovery profile for post-trauma patients. Refreshments served at 3:30 p.m. Hacker Lounge (Neill 216)