Collaborative Research: III: Small: Reconstruction of Diffusion History in Cyber and Human Networks with Applications in Epidemiology and Cybersecurity

网络中的扩散过程可以用来模拟和研究许多真实世界现象，包括在线社交网络上的信息传播，人类网络中的新冠肺炎等传染病，以及互联网上的计算机病毒。非正式地说，扩散历史的重建(RDH)是识别对给定的一组观测提供最佳解释的扩散过程的问题，其中扩散历史是按时间顺序排列的扩散图。本项目的重点是RDH的基本理论和高效的数据驱动算法。RDH的理论和算法在流行病学中识别暴露于病毒的人，在网络安全中跟踪计算机病毒/恶意软件的传播，以及在社交网络中定位泄露的机密信息或谣言的来源和参与者方面具有直接的应用。本项目的第一个部分建立了RDH的部分观察的理论和基本限制，并回答了基本问题，如重建精度和计算复杂性如何与网络规模和数据样本进行衡量。推力2发展了一种新的基于深度学习的算法基础，特别是在具有部分观测的图神经网络和递归神经网络的交叉点上的算法。网络拓扑和时间动力学被嵌入到神经元或神经元的设计以及神经网络的体系结构中。预计开发的算法在准确性、可伸缩性和适用性方面将大大超过最先进的水平。此外，使用合成数据集和真实数据集对理论和算法进行了评估。在该项目下开发的新的深度学习算法及其应用将纳入调查人员教授的课程。该团队继续寻找本科生和来自代表性不足群体的学生参与这一研究项目。这一奖项反映了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。

Diffusion processes in networks can be used to model and study many real-world phenomena, including the spread of information on online social networks, infectious diseases such as COVID-19 in human networks, and computer viruses on the Internet. Informally speaking, reconstruction of diffusion history (RDH) is the problem of identifying a diffusion process that provides the best explanation of a given set of observations, where the diffusion history is a time-sequenced spreading graph. This project focuses on fundamental theories and efficient, data-driven algorithms for RDH. The theories and algorithms for RDH have immediate applications for identifying people exposed to viruses in epidemiology, for tracking the spreading of computer viruses/malware in cyber security, and for locating the sources and participants of leaked classified information or rumors in social networks.Thrust 1 of this project establishes the theories and fundamental limits of RDH with partial observations and answers fundamental questions such as how the reconstruction accuracy and computational complexity scale with network size and data samples. Thrust 2 develops a new algorithmic foundation based on deep learning, especially those at the intersection of graph neural networks and recurrent neural networks, with partial observations. The network topology and temporal dynamics are embedded into the design of cells or neurons and the architecture of the neural networks. The developed algorithms are expected to significantly surpass the state of the art in terms of accuracy, scalability, and applicability. Furthermore, the theories and algorithms are evaluated using both synthetic and real-world datasets. New deep learning algorithms developed under this project and their applications will be integrated into the courses taught by the investigators. The team continues to seek undergraduate students and students from underrepresented groups to involve them in this research project.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.