RI:Small: Relational learning and inference for network models

RI:Small：网络模型的关系学习和推理

• 批准号: 0916443
• 负责人: Yuan Qi
• 金额: $ 35.85万
• 依托单位: Purdue University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2009
• 资助国家: 美国
• 起止时间: 2009-09-01 至 2013-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0916443&HistoricalAwards=false
• 关键词: RI; Small; Relational; learning; inference

Networks are everywhere. Discovering the underlying principles of the networks has great impact on our understanding of complex systems in many scientific, engineering, and social research areas. Nowadays, the availability of network data, such as online social networks from facebook.com or protein-protein interaction data, give researchers unprecedented opportunities to quantitatively study these complex systems. In this project, the PI brings together problems, ideas and techniques from different areas including machine learning, statistics, biology and social sciences, to develop novel computational tools and statistical models for common problems in network inference and learning. The research activities include i) designing nonparametric Bayesian models to discover latent classes from relational data, ii) developing relational Bayesian models, coupled with efficient deterministic approximate inference methods, to predict missing links and node labels, and iii) examining network dynamics at different substructure levels. The developed models, algorithms, and tools for analyzing network data are available to the public via publication and web distribution, disseminating to other machine learning researchers and helping computational biologists and socials scientists analyze massive network data that are being generated with an unprecedented fast speed. The PI incorporates the research results into the graduate-level interdisciplinary courses he teaches and recruits graduate and undergraduate students to conduct research for this project.

网络无处不在。发现网络的基本原理对我们理解许多科学、工程和社会研究领域的复杂系统有很大的影响。如今，网络数据的可获得性，如facebook.com的在线社交网络或蛋白质-蛋白质相互作用数据，为研究人员提供了前所未有的机会来定量研究这些复杂的系统。在这个项目中，PI汇集了不同领域的问题、思想和技术，包括机器学习、统计学、生物学和社会科学，以开发新的计算工具和统计模型来解决网络推理和学习中的常见问题。研究工作包括：i)设计非参数贝叶斯模型以发现关系数据中的潜在类；ii)开发关系贝叶斯模型，结合有效的确定性近似推理方法来预测缺失的链路和节点标签；以及iii)检查不同子结构层次上的网络动态。开发的用于分析网络数据的模型、算法和工具可通过出版物和网络分发向公众提供，传播给其他机器学习研究人员，并帮助计算生物学家和社会科学家分析正在以前所未有的速度生成的海量网络数据。PI将研究成果纳入他教授的研究生水平的跨学科课程中，并招募研究生和本科生为该项目进行研究。