CAREER: Timely Insights: Interpretable, Multi-scale Summarization of Networks over Time

不断发展的网络数据几乎出现在所有学科中。例如，知识或事实通常被构建成知识图，大脑活动通过功能网络表示，神经网络可以被视为不断发展的图结构。该项目旨在开发计算方法和模型，以总结、解释和提供对广泛领域中多个尺度的海量数据（及其潜在动态过程）的见解。专注于知识图谱可以实现设备上和保护隐私的分析（例如，在智能助手上）。神经网络建模有望深入了解其可解释性，并降低其大量的训练计算成本。通过与神经科学专家的合作，这项研究将有助于解码大脑，对精神发育和疾病检测有潜在的影响。这个项目的一个重要部分是将研究与教育相结合的计划。其总体主题是增加计算机和数据科学的多样性，并通过以下方式吸引学生参与图挖掘研究及其现实应用：引入本科生和研究生数据挖掘课程；指导学生进行数据科学项目，造福社会；组织研讨会，吸引不同背景的本科生进入研究生院；组织一个以社交媒体和网络为中心的高中数据科学夏令营，这个主题是对网络科学的成功介绍。网络摘要识别大规模数据中的结构和含义，迄今为止主要集中在非复杂的静态数据上。该项目旨在通过引入总结中的新问题公式（包括以前未被视为图问题的任务）以及理论分析，统一理论和一套新的，可解释的方法和可扩展的算法，弥合网络总结研究与现实世界问题之间的差距。它在不同的尺度上追求三个与网络演化相关的研究任务。在网络规模上，第一个任务侧重于对不断发展的和语义丰富的图数据（例如，异构）进行有效的、监督的或半监督的总结。在多网络尺度上，第二个任务引入了可解释的方法，用于建模和理解不断发展的网络集合及其联合的底层物理过程，这是数据挖掘中尚未研究的问题。通过学术和工业合作，第三项任务探索知识图谱、神经科学、深度神经网络和社会科学的新应用。该项目预计将推进对不断变化的数据进行探索性分析的基础。其成果将通过出版物、教程、研讨会以及开源工具、代码和数据集进行传播。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Evolving network data occur in almost all disciplines. For example, knowledge or facts are often structured into knowledge graphs, brain activity is represented via functional networks, and neural networks can be seen as evolving graph structures. This project aims to develop computational methods and models to summarize, explain, and provide insights into massive data (and their underlying dynamic processes) at multiple scales in a broad range of domains. Focusing on knowledge graphs makes it possible to achieve on-device and privacy-preserving analytics (e.g., on intelligent assistants). Modeling neural networks is expected to give insights into their interpretability and reduce their massive training computational cost. Through collaborations with experts in neuroscience, this research will contribute to decoding the brain, with a potential impact on mental development and disease detection. A significant part of this project is a plan for integrating research with education. Its overarching theme is to increase diversity in computer and data science, and engage students in graph mining research and its real-life applications via: introducing undergraduate and graduate data mining classes; mentoring students on data science projects for social good; organizing a workshop to attract undergraduates from diverse backgrounds to graduate school; and organizing a high-school data science summer camp centered around social media and networks, a theme that is a successful introduction to network science.Network summarization, which identifies structure and meaning in large-scale data, so far has mostly focused on non-complex, static data. This project aims to bridge the gap between network summarization research and real-world problems by introducing novel problem formulations in summarization (including for tasks that have not been previously viewed as graph problems) as well as theoretical analyses, unifying theories, and a suite of new, interpretable methods and scalable algorithms. It pursues three research tasks related to network evolution at different scales. At the network scale, the first task focuses on efficient, supervised or semi-supervised summarization of evolving and semantically-rich graph data (e.g., heterogeneous). At the multi-network scale, the second task introduces interpretable methods for modeling and understanding collections of evolving networks and their joint underlying physical processes, which is an under-studied problem in data mining. Via academic and industrial collaborations, the third task explores new applications in knowledge graphs, neuroscience, deep neural networks, and social sciences. The project is expected to advance the foundations of exploratory analysis of evolving data. Its outcomes will be disseminated through publications, tutorials, workshops, as well as open-source tools, code and datasets.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.