CAREER: New Frontiers in Time Series Analysis

职业：时间序列分析的新领域

• 批准号: 1455172
• 负责人: David Matteson
• 金额: $ 40万
• 依托单位: Cornell University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2015
• 资助国家: 美国
• 起止时间: 2015-07-01 至 2020-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1455172&HistoricalAwards=false
• 关键词: CAREER; New; Frontiers; Time; Series

Big data permeates business, engineering, and science -- the number of connected smart devices, even excluding phones, tablets, and PCs, is projected to grow from billions to tens of billions within five years. Vast data is generated from sensors, GPS, RFID, medical devices, and emergency and energy systems, to provide rich information about untold aspects of the modern world. Despite the ubiquity and significant interest in mining such data, there are few existing analytical tools that are suitable. The investigator focuses on the development of new statistical methodology, extending application of these methods to new fields, and on increasing understanding of the theoretical challenges in data-driven model building and inference. The methods under development have the potential to strengthen research in numerous fields, including astronomy, economics, emergency medical services, neuroscience, and statistics itself.Time series analysis is a rich and historic field, but it remains centered on univariate and low dimensional multivariate analysis. In recent years, big data has begun to permeate business, engineering, and science, but despite the ubiquity and significant interest in mining such data, there are few existing analytical tools suitable for data with a time structured format. The investigator studies the development of new high dimensional and functional time series (HDTS) tools to help researchers and practitioners meet increasingly ambitious inferential and modeling aims. Specifically, the investigator studies: (i) new methods for simultaneous modeling, inference, and forecasting of dynamic functional data; (ii) new structured regularization methods for modeling high dimensional time ordered data; (iii) new adaptive, yet methods of stability analysis for big data monitoring systems; and (iv) linking these new methods with emergent lines of inquiry and providing an infrastructure for answering critical research questions in a wide range of fields.

大数据渗透到商业、工程和科学领域——即使不包括手机、平板电脑和个人电脑，连接的智能设备的数量预计将在五年内从数十亿台增长到数百亿台。传感器、GPS、RFID、医疗设备以及应急和能源系统产生了大量数据，提供了关于现代世界无数方面的丰富信息。尽管对挖掘此类数据的兴趣普遍存在，但很少有现有的分析工具适合。研究者专注于发展新的统计方法，将这些方法扩展到新的领域，并增加对数据驱动模型构建和推理的理论挑战的理解。正在开发的方法有可能加强许多领域的研究，包括天文学、经济学、紧急医疗服务、神经科学和统计学本身。时间序列分析是一个丰富而历史悠久的领域，但它仍然集中在单变量和低维多变量分析上。近年来，大数据已经开始渗透到商业、工程和科学领域，但是，尽管挖掘这些数据无处不在，并且对这些数据有很大的兴趣，但很少有现有的分析工具适合于时间结构化格式的数据。研究者研究了新的高维和功能时间序列（HDTS）工具的发展，以帮助研究人员和从业者满足越来越雄心勃勃的推理和建模目标。具体来说，研究者研究：(i)动态功能数据的同步建模、推理和预测的新方法；（ii）用于高维时间有序数据建模的结构化正则化新方法；（iii）大数据监测系统的新自适应稳定性分析方法；（iv）将这些新方法与新兴的探究线联系起来，并为回答广泛领域的关键研究问题提供基础设施。