CAREER: Towards Exploratory Data Science on Spatio-temporal Big Data

《政府数据公开法案》帮助将数十万个数据集公开提供给科学界和公众；地理空间数据占这些数据的60%以上。该项目描述了构建端到端系统的基础研究，该系统允许数据科学学生和领域科学家交互式地探索时空数据集。首要目标是弥合领域科学家和数据提供者之间的差距。一方面，它帮助不同领域的领域科学家，例如农业、环境科学和政治科学，他们很少有编程技能来探索和访问公开可用的地理空间和时间数据。另一方面，它帮助数据提供者，如政府机构、非营利组织和国家研究实验室，吸引更多的数据科学家开发和利用公共数据。该项目还将开展教育活动，鼓励领域科学家使用公共开放的地理空间数据，这将促进科学结果的可重复性。该项目引入了新的研究方向，旨在建立一个端到端的交互式探索系统，该系统将允许领域科学家处理、分析和可视化pb级的时空数据。它由三个研究部分组成。首先，交互式查询处理器为探索性查询提供实时答案，从而使用户保持参与和活跃，从而提高生产率；这项创新系统地研究了大型地理空间数据的近似查询处理，并将利用深度学习为复杂时空查询的矢量和栅格数据提供准确的误差界限。其次，为了给用户提供一个探索性的界面，一个时空可视化组件提供了一个基于地图的交互式界面，允许用户探索数据的空间和时间属性。当用户通过一个新颖的推荐系统探索大数据集时，这个可视化组件还将为用户提供指导性帮助。最后，随着数据集规模的增长，动态存储系统将不断消耗新的记录，并在分布式存储引擎的基础上更新时空索引、数据摘要和可视化，而分布式存储引擎本身是不可变的，即不支持更新。此外，该项目将建立一个工作原型，科学家可以互动地探索和共享时空数据，并共享公共开放数据。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。

The OPEN government data act helped in making hundreds of thousands of datasets publicly available to the scientific community and the general public; geospatial data comprise over 60% of this data. This project describes basic research towards building an end-to-end system that allows data science students and domain scientists to interactively explore spatio-temporal datasets. The overarching goal is to bridge the gap between domain scientists and data providers. On one end, it helps domain scientists in various fields, e.g., agriculture, environmental science, and political science, who have little programming skills to explore and access publicly available geospatial and temporal data. On the other end, it helps data providers, e.g., government agencies, non-profit organizations, and national research labs, to attract more data scientists to exploit and utilize public data. The project will also establish educational activities that encourage domain scientists to use public open geospatial data which will promote the reproducibility of scientific results.This project introduces new research directions that are geared towards building an end-to-end interactive exploratory system that will allow domain scientists to process, analyze, and visualize petabytes of spatio-temporal data. It consists of three research components. First, an interactive query processor provides a real-time answer to exploratory queries so that the user will stay engaged and active which increases the productivity; this innovation systematically studies approximate query processing for large geospatial data and will utilize deep learning to provide accurate error bounds for both vector and raster data for complex spatio-temporal queries. Second, to provide users with an exploratory interface, a spatio-temporal visualization component provides an interactive map-based interface that allows users to explore the spatial and temporal attributes of the data. This visualization component will also provide guided assistance to users when exploring big datasets through a novel recommendation system. Lastly, as the datasets grow in size, a dynamic storage system will continuously consume the new records and update the spatio-temporal indexes, data summaries, and visualizations on top of a distributed storage engine which is inherently immutable, i.e., does not support updates. Additionally, this project will build a working prototype where scientists can interactively explore and share spatio-temporal data and share public open 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.