CDS&E: Graph-Based Learning and Uncertainty Quantification for Large-Scale Complex Data

点击翻译按钮获取中文摘要

Many modern empirical and experimental researches routinely produce massive amounts of high-dimensional and complex data. Graphical model is one of the most important tools to visualize and exploit hidden, latent, low-dimensional structure in high-dimensional complex data. Motivated by neuroscience and genomics applications, the research project aims to develop cutting-edge graphical model based computational and statistical methods to uncover latent network structures in large-scale complex data. To enhance the reproducibility of scientific studies, the project provides rigorous tools on uncertainty quantification. The software packages will be developed to implement the proposed methods. The research project focuses on theoretical and methodological development for graphical models in high-dimensional complex data arising from neuroscience and genomics applications. Specific projects include proposing clustering based latent variable models for dimensional reduction with a large number of variables, optimal distributed learning and inference for massive heterogenous data under communication constraint, and exponential family random effect graphical models for high-dimensional longitudinal data under complex dependence. The proposed models are equipped with a powerful multiple testing procedure to control false discovery rate in a rigorous manner. The computational algorithms will be developed in each application.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.