EAGER: Science in the Time of Big Data

科学界充斥着“大数据”，但很少有实践生态学家使用这些数据来回答重要的生态问题。相反，他们依靠传统的方法收集新的实验数据，这些数据集中在特定的物种、栖息地或问题上。此外，分析大数据集所需的数据密集型计算方法对大多数研究人员来说并不容易获得。这个高风险、高回报的项目可能会极大地改变生态学家解决问题的方式和他们解决的问题类型。因此，它代表了一个重大贡献，NSF的努力，扩大生态研究的新方向，提供答案，更复杂的问题。一个知识驱动的开放获取系统，随着数据流的种类和规模的增加，它会“学习”并变得更有效和更容易使用，这是在大数据时代及时取得科学进步所必需的。这种方法的核心是建立数据库和基于假设的调查，导致在推导新的或完善的假设作为一个结果，改善访问动态数据库之间的联系。研究人员最近实施了一种假设驱动的、基于过程的分析方法，该方法在概念上与数据密集型机器学习方法相结合。这种综合方法使他们能够使用多个长期数据集将一系列不同的机械解释缩小到一个最有可能的过程。然后通过短期实验对这一过程进行测试，节省了时间和金钱，并得出了比上述更传统的方法更明确的答案。为了进一步推进这一方法，该项目将测试，完善和自动化这一新的综合努力，以开发一个能够显着推进环境科学的原型网络基础设施。开放获取数据、编程脚本和衍生数据产品将减少知识从个人转移到研究社区的时滞，可能会提高科学进步的速度，并为如何处理质量参差不齐的大量数据提供过滤器和记忆。一名博士后研究员将与来自三所大学（新墨西哥州州立大学、德克萨斯大学埃尔帕索分校和万年青学院）和一家公司（微软）的计算机科学家、生态学家和生态信息学专家合作开发、测试和自动化这个知识学习分析系统。将举办两次讲习班，测试该系统在使用各种数据集的同时进行学习的能力，并向各种用户介绍这一方法。

The scientific community is awash in 'big data' but few practicing ecologists use these data to answer important ecological questions. They rely instead on the traditional approach of collecting new, experimental data focused on particular species, habitats, or problems. In addition, the data-intensive computational methods commonly needed to analyze big datasets are not easily accessible to most researchers. This high-risk, high-reward project could dramatically alter both the ways in which ecologists address questions and the types of questions that they tackle. It therefore represents a major contribution to NSF's efforts to extend ecological research in new directions to provide answers to more complex questions. A knowledge-driven, open access system that 'learns' and becomes more efficient and easier to use as data streams increase in variety and size is needed for timely scientific progress in an era of big data. This approach is centered on establishing linkages between databases and hypothesis-based inquiry that result in the derivation of new or refined hypotheses as a result of improved access to dynamic databases. The investigators recently implemented a hypothesis-driven, process-based analytical methodology that was conceptually integrated with a data-intensive machine learning approach. This integrated approach allowed them to use multiple long-term datasets to narrow a diverse suite of mechanistic explanations to a single, most likely process. This process was then tested by a short-term experiment that saved time and money and yielded a more definitive answer than the more traditional approach described above. To further this approach, this project will test, refine, and automate this new integrative effort to develop a prototype cyber-infrastructure capable of significantly advancing the environmental sciences. Open access data, programming scripts, and derived data products will reduce the time lag for knowledge transfer from an individual to the research community, likely increase the speed of scientific progress, and provide a filter and memory for how to deal with large amounts of data of mixed quality. A postdoctoral researcher will work collaboratively with computer scientists, ecologists, and eco-informatics experts from three universities (New Mexico State University, University of Texas El Paso, and Evergreen College) and one corporation (Microsoft) to develop, test, and automate this knowledge-learning analytics system. Two workshops will be organized to test the ability of the system to learn while using diverse datasets and to introduce the approach to a wide variety of users.