CAREER: Inference-Driven Data Processing and Acquisition: Scalability, Robustness and Control

无处不在的传感技术使新的服务和应用成为可能，这些服务和应用涉及我们生活的方方面面。然而，随之而来的数据洪流是我们解释，处理和存储数据的能力面前的砖墙。此外，收集的数据并非都是信息性的，也不是与我们要完成的任务相协调的。该项目介绍了数据处理的新方法，有望在处理大量数据时带来惊人的加速，并探索了许多推理问题的原则性控制数据采集范例。研究活动预计将推进民用基础设施，医疗保健，能源和交通等新兴网络物理系统中数据处理和获取的理论和实践。研究涉及1）开发和分析基于子空间的数据处理方法的基本原理，这些方法同时具有可扩展性和对结构保持数据草图中的离群值的鲁棒性，以及对于保持底层低维数据结构的变换不变的数据子空间公式，（二）探索控制数据采集的渐近制度，大量的观察与放松的概念，最优解开最优设计的完整结构，并认识到统一的原则，设计有效的控制政策，为主机的控制推理问题研究了基本结果在工程结构损伤检测和表征的结构健康监测中的实际意义。教育活动包括建立一个新的微型传感实验室，开发一个关于从大数据和序列分析中学习的入门课程，编写一个直观教程语料库，阐述研究成果的核心概念，以及指导高级设计项目。

Ubiquitous sensing has enabled new services and applications that markevery aspect of our lives. However, the ensuing data deluge is a brickwall in face of our abilities to interpret, process and store data. Also, notall the data collected is informative, nor well-attuned to the tasks wecare to accomplish. This project introduces new approaches for dataprocessing that hold promise to bring about stunning speedups in theprocessing of massive data, and explores principled controlled dataacquisition paradigms for numerous inference problems. The researchactivities are expected to advance the theory and practice of dataprocessing and acquisition in emerging cyber-physical systems for civilinfrastructure, healthcare, energy, and transportation.The research involves 1) developing, and analyzing the fundamentallimits of, transformative subspace-based approaches to data processingthat are simultaneously scalable and robust to outliers using subspacepursuit in structure-preserving data sketches, and data-subspaceformulations which are invariant to transformations that preserve theunderlying low-dimensional data structures, 2) exploring controlled dataacquisition in asymptotic regimes of large number of observations withrelaxed notions of optimality to unravel the complete structure ofoptimal design and recognize unifying principles for the design ofefficient control policies for a host of controlled inference problems. Thepractical implications of the fundamental results are studied in thecontext of structural health monitoring for damage detection andcharacterization of engineering structures. The educational activitiesinclude establishing a new miniature sensing lab, developing anintroductory course on learning from big data and sequential analysis,compiling a corpus of intuitive tutorials laying out the core concepts ofresearch results, and mentoring of senior design projects.