CAREER: Model-based Analysis of Dynamic Networks using Continuous-time Network Models

网络以多种形式存在于我们周围，从在线社交网络到公共交通网络，再到生物学中的基因网络。大多数网络随着时间的推移而变化，通常称为时态网络或动态网络。在这个项目中，将开发一个用于建模和分析动态网络的框架，该网络随着时间的推移不断变化，即使网络可能只被周期性地观察到。这一框架通过开发模型将网络的基本动态与观察网络的时间分开，从而推动了网络科学的跨学科领域以及计算机和信息科学的发展。该框架可用于分析许多科学学科和公共卫生应用中的动态网络数据，包括人与人之间面对面互动的网络，这可以帮助科学家更好地了解新冠肺炎等传染病的传播。该项目通过为本科生和研究生创建动态网络教学课程来推进网络科学教育。该项目还培训新的研究生和本科生，包括托莱多大学ACM-W分会的女学生，进行跨学科的数据科学研究。最后，该项目开发了分析动态网络的方法，并将其集成到开源的DyNetworkX Python包中，以便其他人能够以有效的方式使用它们。众所周知，网络中的临时动力学提供了关于网络建模的底层复杂系统的关键信息。虽然在理解静态网络的结构方面已经取得了重大进展，但通常通过创建在某一任意时间段上聚集的离散时间快照来以特别的方式结合动态，主要是为了便于分析。该项目的目标是开发一个统一的框架，用于使用连续时间模型对动态网络进行基于模型的分析，该模型既可以应用于离散动态网络数据，也可以应用于连续时间动态网络数据。为了实现这一目标，研究小组将瞄准五个具体目标：1)从随时间变化的关系事件的聚合计数中学习连续时间网络模型；2)为具有持续时间的时间戳事件创建基于Hawkes过程的生成模型；3)开发用于分析动态网络的核平滑方法；4)对动态网络数据中不同类型的测量误差进行建模；以及5)创建时间和内存高效的动态图形数据结构，以实现对大型动态网络的高时间分辨率分析。在目前的网络科学课程和教科书中，网络动力学的覆盖面很小。本项目将开发的基于模型的分析技术建立在基本网络理论和对真实网络的经验观察基础上，因此是整合到典型的研究生或本科生网络科学课程中的理想选择。研究人员将为动态网络表示、模型和分析方法开发可公开提供的课程。这个项目的结果将提供一个连续时间网络模型实现的可能性的一瞥，并指导未来在动态网络上的研究和教育工作。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Networks are all around us in many forms, ranging from online social networks to public transportation networks to gene networks in biology. Most networks change over time and are often called temporal or dynamic networks. In this project, a framework for modeling and analyzing dynamic networks that change continuously over time will be developed, even though the networks may only be periodically observed. This framework advances the interdisciplinary field of network science along with the computer and information sciences by developing models to separate the underlying dynamics of the networks from the times at which the networks are observed. The framework can be applied to analyze dynamic network data in many scientific disciplines and in public health applications, including networks of face-to-face interactions between people, which can help scientists better understand the spread of infectious diseases such as COVID-19. This project advances education in network science by creating a curriculum for instruction of dynamic networks at the undergraduate and graduate levels. The project also trains new graduate and undergraduate students, including female students from the University of Toledo's ACM-W chapter, in interdisciplinary data science research. Finally, the project develops and integrates methods for analyzing dynamic networks into the open-source DyNetworkX Python package to reach others who could use them in impactful ways.Temporal dynamics in networks are known to provide crucial information about the underlying complex systems being modeled by the networks. While significant advances have been made towards understanding the structure of static networks, dynamics are usually incorporated in an ad-hoc manner by creating discrete time snapshots aggregated over some arbitrary time period, primarily for convenience of analysis. The goal of this project is to develop a unified framework for model-based analysis of dynamic networks using continuous-time models that can be applied to both discrete- and continuous-time dynamic network data. Towards this goal, the research team will target five specific aims: 1) learning continuous-time network models from aggregated counts of relational events over time, 2) creating Hawkes process-based generative models for timestamped events with durations, 3) developing kernel smoothing approaches for analyzing dynamic networks, 4) modeling different types of measurement error in dynamic network data, and 5) creating time- and memory-efficient dynamic graph data structures to enable analysis of large dynamic networks with high temporal resolution. Dynamics of networks are given minimal coverage in current network science curricula and textbooks. The model-based analysis techniques to be developed in this project build upon fundamental network theory and empirical observations about real networks and are thus ideal for integration into a typical graduate or undergraduate network science course. The investigator will develop a publicly-available curriculum for instruction on dynamic network representations, models, and analysis methods. The results of this project will provide a glimpse of the possibilities enabled by continuous-time network models and guide future research and education efforts on dynamic networks.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.