Collaborative Research: New Statistical Learning for Complex Heterogeneous Data

该项目侧重于医学成像和社交网络中产生的复杂异构数据的几个重要和具有挑战性的问题。主要目标是开发强大而创新的统计机器学习方法和工具，能够灵活地模拟图像之间的信号异质性，将成像数据与多模态和空间分布数据集成，并解决网络数据的异质性。研究和教育的综合方案将在许多不同的领域产生重大影响，如生物医学研究、基因组研究、环境研究、公共卫生研究、社会和政治科学等。该项目还将促进跨学科研究和来自不同领域的科学家的合作。该项目将通过利用来自一般人群的个体差异，以及整合多种来源的成像信息来提高疾病诊断和治疗结果的预测准确性，从而在异质性学习和建模方面取得实质性进展。此外，本项目通过利用节点协变量信息来分析异构网络数据，开发了创新的无监督学习方法。研究计划的每个组成部分都包含广泛的主题，从方法论和计算开发到现实世界问题的应用。具体而言，pi研究了主题变量的图像上标量回归模型，以纳入脑成像数据的异质性变化，乳腺癌成像数据的多维张量学习方法，网络数据的灵活高斯图模型，以及网络连接的异构数据的新型聚类框架。此外，先进的优化技术、算法和计算技术的发展将适用于由大规模异构数据引起的许多实际问题。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This project focuses on several important and challenging issues concerning complex heterogeneous data that arise from medical imaging and social networks. The major goals are to develop powerful and innovative statistical machine learning methods and tools that are able to flexibly model signal heterogeneity across images, integrate imaging data with multimodal and spatially distributed data, and tackle heterogeneity of network data. The integrated program of research and education will have significant impacts in many different fields such as biomedical studies, genomic research, environmental studies, public health research, and social and political sciences, among others. Te project will also stimulate interdisciplinary research and collaboration with scientists from disparate fields.This project will lead to substantial advancement in heterogeneity learning and modeling through exploiting individual variation from the general population, and integration of multiple sources of imaging information to enhance prediction accuracy for disease diagnoses and treatment outcomes. In addition, this project develops innovative unsupervised learning methods through utilizing node covariate information for analyzing heterogeneous network data. Each component of the research plan contains a broad range of topics, from methodological and computational development to applications in real world problems. Specifically, the PIs study subject-variant scalar-on-image regression models to incorporate the heterogeneity variation for brain imaging data, multi-dimensional tensor learning methods for breast cancer imaging data, flexible Gaussian graphical models for network data, and a novel clustering framework for heterogeneous data that are linked by networks. Furthermore, the development of advanced optimization techniques, algorithms and computational technologies will be applicable to many practical problems arising from large-scale heterogeneous data.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.