Collaborative Research: Learning Classifiers From Autonomous, Semantically Heterogeneous, Distributed Data

网络、传感器、存储、计算和高吞吐量数据采集方面的进步，导致了人类活动许多领域中自主、分布式数据源的激增。我们发现、共享、整合和分析不同类型数据的能力，正推动着生物、物理、社会科学和工程领域的新发现。基于统计的机器学习算法为从数据中发现实验可测试的预测模型和假设提供了一些最具成本效益的方法。然而，数据源的庞大规模、分布式性质和自主性（以及随之而来的访问、允许的查询、处理能力、结构、组织以及底层数据模型和数据语义方面的差异）为有效利用机器学习带来了障碍。本研究旨在通过开发高效、资源感知的分布式算法和软件服务来克服这些障碍，以支持协作、集成的知识获取。研究团队将在爱荷华州立大学和堪萨斯州立大学的网络数据库和服务的分布式测试平台上使用基准数据集、相关数据模型和本体以及用户指定的本体间映射来实现、部署和评估结果算法。由此产生的开源软件可以潜在地改变协同电子科学，就像网络改变了信息共享一样。这项研究的更广泛的影响包括为研究生和本科生提供更多的以研究为基础的培训机会，跨学科合作，代表性不足的群体的参与，以及越来越复杂的软件的开发，以支持协作，综合电子科学。项目网站（http://www.cild.iastate.edu/projects/indus.html）提供了对项目信息、基准数据、出版物、软件和文档的访问。

Advances in networks, sensors, storage, computing, and high throughput data acquisition, have led to a proliferation of autonomous, distributed data sources in many areas of human activity. New discoveries in biological, physical, and social sciences and engineering are being driven by our ability to discover, share, integrate and analyze disparate types of data. Statistically-based machine learning algorithms offer some of the most cost-effective approaches to discovery of experimentally testable predictive models and hypotheses from data. However, the large size, distributed nature, and autonomy of the data sources (and the attendant differences in access, queries allowed, processing capabilities, structure, organization, and underlying data models and data semantics) present hurdles to effective utilization of machine learning. This research aims to overcome these hurdles by developing efficient, resource-aware distributed algorithms and software services to support collaborative, integrative knowledge acquisition such a setting. The research team will implement, deploy, and evaluate the resulting algorithms using benchmark data sets, associated data models and ontologies, and user-specified inter-ontology mappings on a distributed test-bed of networked databases and services at Iowa State University and Kansas State University. The resulting open-source software can potentially transform collaborative e-science in the same way that Web has transformed information sharing. Broader impacts of this research include enhanced opportunities for research-based training of graduate and undergraduate students, interdisciplinary collaborations, participation of under-represented groups, and development of increasingly sophisticated software to support collaborative, integrative e-science. The project web site (http://www.cild.iastate.edu/projects/indus.html) provides access to information about the project, benchmark data, publications, software, and documentation.