III: Medium: Collaborative Research: Scaling Machine Learning to Massive Datasets---A Logic Based Approach

III：媒介：协作研究：将机器学习扩展到海量数据集——基于逻辑的方法

• 批准号: 1302690
• 负责人: Neoklis Polyzotis
• 金额: $ 33.3万
• 依托单位: University of California-Santa Cruz
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2013
• 资助国家: 美国
• 起止时间: 2013-09-01 至 2017-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1302690&HistoricalAwards=false
• 关键词: III; Medium; Collaborative; Research; Scaling

Machine learning (ML) algorithms have become ubiquitous across applications as diverse as science, engineering, business, finance, education and healthcare. However, development of ML software that can scale to massive datasets and that are also easy-to-use remains a challenge in part due to the fact that developing an ML tool currently requires the implementation of a deep software stack, from the actual runtime (i.e., how an ML algorithm is executed) to the API exposed to the users.This project aims to develop DeML, a system to support the authoring and execution of ML tools. Specifically, DeML would allow ML algorithms to be formulated in the form of a declarative query over the training dataset. DeML optimizes the execution of the query over a computing platform (e.g., Amazon EC2 or SQL Azure), taking into account the characteristics of the algorithm, the data, and the available computational resources. Adoption of DeML would greatly reduce the effort required to develop scalable implementations of ML algorithms. The project is organized around three thrusts: (i) Development of a declarative query language, based on extensions of Datalog; (ii) Analysis of runtime of DeML queries; (iii) Optimization of dataflow of DeML queries based on the characteristics of data sources and the capabilities of the underlying execution platform. The resulting open source DeML prototype implementation will be made freely available to the community through the project web page at: http://deml.cs.ucla.edu.The availability of the DeML could greatly lower the effort needed to author scalable implementations of ML algorithms for analysis of massive datasets, which in turn would increase the availability of such tools to the broader community. Experience gained by implementing and deploying ML algorithms at scale over modern cloud-computing platforms, could help inform critical design choices in the development of future cloud computing platforms for big data analytics, and hence impact a broad range of scientific, engineering, national security, healthcare and business applications of big data analytics. The project offers enhanced opportunities for research-based advanced training of graduate and undergraduate students, including members of groups that are currently under-represented in computer science, in databases, machine learning, and cloud computing.

机器学习（ML）算法在科学、工程、商业、金融、教育和医疗保健等各种应用中无处不在。然而，开发可以扩展到大量数据集并且易于使用的ML软件仍然是一个挑战，部分原因是开发ML工具目前需要实现深度软件堆栈，从实际运行时（即ML算法如何执行）到向用户公开的API。该项目旨在开发DeML，一个支持ML工具的创作和执行的系统。具体来说，DeML将允许ML算法以对训练数据集的声明性查询的形式进行公式化。DeML在计算平台（例如Amazon EC2或SQL Azure）上优化查询的执行，同时考虑到算法、数据和可用计算资源的特征。采用DeML将大大减少开发ML算法的可伸缩实现所需的工作量。该项目有三个重点：(i)根据Datalog的扩展开发一种声明性查询语言；（ii）分析DeML查询的运行时；（iii）根据数据源的特点和底层执行平台的能力，优化DeML查询的数据流。由此产生的开源DeML原型实现将通过项目网页：http://deml.cs.ucla.edu.The免费提供给社区，DeML的可用性可以大大降低编写用于分析大量数据集的ML算法的可扩展实现所需的工作量，这反过来将增加此类工具对更广泛社区的可用性。通过在现代云计算平台上大规模实施和部署机器学习算法所获得的经验，可以帮助为未来大数据分析云计算平台开发中的关键设计选择提供信息，从而影响大数据分析的广泛科学、工程、国家安全、医疗保健和商业应用。该项目为研究生和本科生提供了更多基于研究的高级培训机会，包括目前在计算机科学、数据库、机器学习和云计算领域代表性不足的小组成员。