EAGER: Developing a temporally adaptive decomposition framework for analyzing long-term echosounder time series

EAGER：开发用于分析长期回声测深仪时间序列的时间自适应分解框架

基本信息

批准号：
1849930
负责人：
Wu-Jung Lee
金额：
$ 28.16万
依托单位：
University of Washington
依托单位国家：
美国
项目类别：
Standard Grant
财政年份：
2019
资助国家：
美国
起止时间：
2019-01-01 至 2021-12-31
项目状态：
已结题

来源：
https://www.nsf.gov/awardsearch/showAward?AWD_ID=1849930&HistoricalAwards=false
关键词：
EAGER Developing temporally adaptive decomposition

项目摘要

Scientific echosounders are high-frequency active sonar systems used extensively to study life in the ocean, especially the least-characterized mid-trophic animals, such as zooplankton and fish. The recent surge in the deployment of autonomous echosounders has resulted in a deluge of echo data from a variety of ocean observing platforms, including vehicles and moorings. These data sets harbor great potential for advancing our understanding of marine ecosystems at spatial and temporal scales never before possible. However, the volume and complexity of the data make efficient and effective analysis a major challenge. This research is aimed at establishing a data-driven methodology to automatically extract features from echosounder data, thus unblocking this bottleneck between instrumentation capacity (to collect large data) and interpretation capability (to analyze large data). The developed method will expedite the currently labor-intensive data analysis process and help extract synoptic information from large echo data sets. Results of this research will be disseminated through open-source echo analysis software packages and online educational materials created as integrated components of this project. This research contributes directly to improving our ability to better understand the marine ecosystems, which are critical for global biodiversity and the economic well-being of a significant population of society.This research will develop a method for automated discovery and tracking of low-dimensional spatio-temporal structures from high-dimensional echo data by adapting a state-of-the-art dynamic non-negative matrix factorization (NMF) formulation. NMF is an unsupervised machine learning technique that decomposes complex data into a linear combination of a smaller set of quantitative descriptors that are more tractable and interpretable than the original data. This technique has found great success in many basic research and applied fields, including computer vision, neuroscience, natural language processing, and recommender systems for internet retailers. In the context of echo analysis, these low-dimensional descriptors characterize the vertical movements and grouping activities of marine organisms in the water column imaged by the echosounder. However, traditional NMF does not account well for a changing set of phenomena in the data. Therefore, this project aims to develop a method that can provide a temporally adaptive decomposition in order to accommodate the complexity and high temporal variability of transient features in long-term echosounder data. The researchers will use data collected by the two cabled Endurance Array echosounders in NSF?s Ocean Observatories Initiative (OOI) as a test bed for methodology development. The stability, robustness, and interpretability of the developed method will be evaluated and compared with conventional echo analysis routines. This effort will lay the foundation for further machine learning and statistical studies of large echo datasets from the OOI and other ocean observing systems.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.

科学回声器是高频活跃的声纳系统，用于研究海洋中的生命，尤其是特征最少的中营养动物，例如浮游动物和鱼类。自动回声器的部署最近部署的激增导致了来自各种海洋观察平台的回声数据，包括车辆和系泊设备。这些数据集具有巨大的潜力，可以促进我们对空间和时间尺度上从未做出的空间和时间尺度的理解。但是，数据的数量和复杂性使高效分析成为主要挑战。这项研究旨在建立一个数据驱动的方法，以自动从Echosounder数据中提取特征，从而解除仪器能力（收集大数据）和解释能力（分析大数据）之间的瓶颈。开发的方法将加快目前的劳动密集型数据分析过程，并有助于从大回声数据集中提取概要信息。这项研究的结果将通过开源回声分析软件包和作为该项目集成组件创建的在线教育材料进行传播。 This research contributes directly to improving our ability to better understand the marine ecosystems, which are critical for global biodiversity and the economic well-being of a significant population of society.This research will develop a method for automated discovery and tracking of low-dimensional spatio-temporal structures from high-dimensional echo data by adapting a state-of-the-art dynamic non-negative matrix factorization (NMF) formulation. NMF是一种无监督的机器学习技术，将复杂的数据分解为一组较小的定量描述符的线性组合，这些定量描述符比原始数据更可拖动和可解释。该技术在许多基础研究和应用领域中都取得了巨大的成功，包括计算机视觉，神经科学，自然语言处理以及互联网零售商的推荐系统。在回声分析的背景下，这些低维描述符表征了回声器成像的水柱中海洋生物的垂直运动和分组活动。但是，传统的NMF无法很好地解释数据中不断变化的现象。因此，该项目旨在开发一种可以提供时间自适应分解的方法，以适应长期回声器数据中瞬态特征的复杂性和高时间变化。研究人员将使用NSF海洋天文台倡议（OOI）中两个有线耐力阵列收集的数据作为方法论开发的测试床。将评估开发方法的稳定性，鲁棒性和可解释性，并将其与常规回声分析程序进行比较。这项努力将为进一步的机器学习和统计研究奠定基础，从OOI和其他海洋观察系统中进行大型回声数据集。该奖项反映了NSF的法定任务，并被认为是值得通过基金会的知识分子和更广泛影响的评估评估来支持的。