SHF: Small: Towards High-Performance Machine Learning on Graphs

通过社交网络、智能手机，以及更普遍的网络物理系统，世界变得越来越相互交织。互联的人员、设备和系统带来了一个紧迫的研究挑战，即需要了解动态网络中不同实体之间的关系。由于带有顶点和边的图是这种关系的自然表示，在这些图(网络)上的机器学习对于从生物、金融到社会科学的各种应用都是有益的。重要的是，这些网络的规模呈指数级增长。因此，不仅需要存储和管理大规模的、动态的图形数据集，而且需要支持机器学习任务的快速执行。本项目设计和开发了一套新的针对大图数据集的高性能机器学习算法和系统。研究目标是通过在高性能计算系统上实现基于图形的机器学习的系统级创新，以及开发用于图形提取和学习的新的可扩展、健壮的算法来实现的。新的图形学习系统将显著提高图形机器学习的水平，为从社会科学到生命科学的一系列科学学科提供新的工具。一些本科生和研究生，包括来自代表性不足群体的学生，将参加这一项目。研究成果将作为开源项目以及在主要会议和科学期刊上的技术出版物进行传播。这一奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

The world has become increasingly intertwined through social networks, smart phones, and more generally cyber-physical systems. Connected people, devices, and systems introduce a pressing research challenge, that is, the need of understanding the relationships between different entities in dynamic networks. As graphs with vertices and edges are a natural representation of such relationships, machine learning on these graphs (networks) is beneficial to a variety of applications spanning from biology, finance, to social science. Importantly, these networks grow exponentially in size. As a result, one needs to not only store and manage large-scale, dynamic graph datasets, but also support fast execution of machine-learning tasks.This project designs and develops a new set of high-performance machine-learning algorithms and systems for big graph datasets. The research goal is achieved through systems-level innovations for enabling graph-based machine learning on high-performance computing systems, coupled with the development of new scalable, robust algorithms for graph extraction and learning. The new graph learning system will significantly advance the state of the art in machine learning on graphs, delivering new tools to an array of scientific disciplines from social to life sciences. A number of undergraduate and graduate students, including those from under-represented groups, will participate in this project. The research results will be disseminated as open-source projects, as well as technical publications in major conferences and scientific journals.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.