Big Data Fusion

数据融合是一种机制，用于以一致的方式收集和组合从多个来源获取的数据，以及时获得更可靠、更少体积的处理数据。它通常用于涉及多个传感器的许多领域，例如国防和安全，医疗诊断和手术，智能交通系统。随着我们进入数据泛滥的时代，不仅数据量大大增加，而且存在各种类型的数据源，例如传感器测量和记录以及网络空间（例如社交网络）。结构化和非结构化数据的这种令人难以置信的增长和可用性导致了数量，速度和多样性方面的问题，称为大数据。然而，目前的数据融合技术无法真正应对大数据带来的挑战。在这个项目中，我们提出了一个多学科的大数据融合方法。我们将结合计算机科学中的联合收割机近似理论，物理学中的复杂性理论，以及信号处理中的分布式和稀疏系统来开发融合大量数据的新方法。特别是，我们将集中在数据融合的三个基本组成部分：注册，关联和融合。我们建议使用系统稀疏性和运动动态性来开发一种新的配准模型，以对齐大量的设备。为了关联或关联数据，我们将开发一个近似的多帧最近邻数据关联来加速关联过程，并使用复杂网络来提取网络空间中的深度相关性以确定诚实的关系。由于大数据采用多种格式，我们建议开发基于一阶和二阶统计的通用数据融合，以提供接近最佳的数据融合。拟议的研究将集中在三个重要行业的应用：能源，无线通信和环境。这项研究不仅开发了大数据的基础数据和信号处理算法，而且还使数据融合适用于最终涉及大数据的领域。大数据产业价值超过1000亿美元，每年增长10%。到2018年，仅美国在这方面就可能面临约17万人的短缺。拟议的研究将为快速增长的市场培养人才。**

Data fusion is a mechanism for the data acquired from multiple sources to be gathered and combined in a consistent way to obtain more reliable, less voluminous processed data in a timely manner. It is conventionally used in many sectors involving multiple sensors, such as defense and security, medical diagnostic and surgery, intelligent transportation systems. As we are moving into an era of data deluge, not only is the amount of data greatly increasing, but there is a wide variety of types of data sources, such as sensor measurements and recordings, and cyberspace (e.g. social networks). This incredible growth and availability of both structured and unstructured data causing problems in volume, velocity and variety is termed big data. Current data fusion techniques, however, cannot really handle the challenges posted by big data.****In this project, we propose a multi-disciplinary approach for big data fusion. We will combine approximation theory in computer science, complexity theory in physics, and distributed and sparse systems in signal processing to develop novel methods for fusing massive amounts of data. In particular, we will focus on the three basic components of data fusion: registration, association and fusion. We propose using system sparseness and motion dynamic to develop a new registration model to align a large amount of devices. To associate or correlate the data, we will develop an approximate multi-frame nearest neighbor data association to speed up the association process and use complex networks to extract the deep correlations in the cyberspace to determine honest relationships. Since big data takes on a variety of formats, we propose to develop a universal data fusion based on first- and second- order statistics to provide a close-to-optimal data fusion.***The proposed research will focus on applications in three important industries: energy, wireless communications and the environment. The proposed research does not only develop fundamental data and signal processing algorithms for big data, but it also makes data fusion applicable to areas that will eventually involve big data. The big data industry is worth more than $100 billion and growing at 10% a year. By 2018, the United States alone could face a shortage of about 170,000 people in this area. The proposed research will prepare talented people for the rapidly growing market.**