EAGER: Asynchronous Event Models for State-Topology Co-Evolution of Temporal Networks

EAGER：时态网络状态拓扑协同演化的异步事件模型

• 批准号: 1639792
• 负责人: Duen Horng Chau
• 金额: $ 20万
• 依托单位: Georgia Tech Research Corporation
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-07-15 至 2019-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1639792&HistoricalAwards=false
• 关键词: EAGER; Asynchronous; Event; Models; State

The purpose of this research project is to develop probabilistic models and the related machine learning algorithms for modeling network evolution and dynamics. The research lays theoretic foundations and provides practical tools for scientists to control networks in order to achieve desirable outcomes. Although the research is widely applicable, the research team primarily considers two application areas: social networks and P2P microfinance. In social networks, this project brings practical values to the Internet industry by better understanding and modeling of user behaviors and their impacts on social ties and social group formation. For P2P microfinance, this project has the potential to better engage not-for-profit lenders and thus to help small business in developing countries. Furthermore, the research provides materials and contents for both undergraduate and graduate education and helps students develop interdisciplinary mindsets and tools needed to tackle real-world problems. This proposed research aims to develop machine learning theory and algorithms for networked asynchronous and interdependent event streams arising from modern applications. The researchers especially emphasize methodology that can handle temporal networks when the underlying network structures are undergoing substantial changes. One major theme of the proposal is the modeling of the interplay between network node dynamics and network topology dynamics, or network co-evolution. The researchers propose a novel framework based on multivariate point processes for modeling and analyzing event data. The methods significantly expand the application area of conventional machine learning techniques. One example is to answer the question ``who will do what and when'', which is critical to event sequence modeling in network data analysis where traditional machine learning algorithms are difficult to apply.

该研究项目的目的是开发概率模型和相关的机器学习算法，用于建模网络演化和动态。该研究为科学家控制网络以达到预期效果奠定了理论基础，并提供了实用工具。虽然该研究具有广泛的适用性，但研究团队主要考虑两个应用领域：社交网络和P2P小额信贷。在社交网络中，该项目通过更好地理解和建模用户行为及其对社会关系和社会群体形成的影响，为互联网行业带来了实用价值。对于P2P小额信贷，该项目有可能更好地吸引非营利贷款人，从而帮助发展中国家的小企业。此外，该研究为本科和研究生教育提供了材料和内容，并帮助学生发展解决现实问题所需的跨学科思维方式和工具。 这项研究旨在为现代应用中产生的网络异步和相互依赖的事件流开发机器学习理论和算法。研究人员特别强调，当底层网络结构发生重大变化时，可以处理时间网络的方法。该提案的一个主要主题是对网络节点动态和网络拓扑动态之间的相互作用或网络协同进化进行建模。研究人员提出了一种基于多变量点过程的新框架，用于建模和分析事件数据。这些方法显着扩展了传统机器学习技术的应用领域。一个例子是回答“谁将在何时做什么”的问题，这对于传统机器学习算法难以应用的网络数据分析中的事件序列建模至关重要。

项目成果

DyRep: Learning Representations over Dynamic Graphs

• 发表时间: 2019-05
• 作者: Rakshit S. Trivedi;Mehrdad Farajtabar;P. Biswal;H. Zha

Decoupled Learning for Factorial Marked Temporal Point Processes

• DOI: 10.1145/3219819.3220035
• 发表时间: 2018-01
• 期刊: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining
• 作者: Weichang Wu;Junchi Yan;Xiaokang Yang;H. Zha

Learning Conditional Generative Models for Temporal Point Processes

• DOI: 10.1609/aaai.v32i1.12072
• 发表时间: 2018-04
• 作者: Shuai Xiao;Hongteng Xu;Junchi Yan;Mehrdad Farajtabar;Xiaokang Yang;Le Song;H. Zha