Collaborative Research: Graph Learning, Generation, and Optimization with Highly Expressive Graph Neural Network Models

In today's data-driven world, many complex systems can be represented as interconnected networks or graphs. This project aims to develop new methods for analyzing, generating, and optimizing these graph structures, with potential applications in areas such as social network analysis and molecular design. By improving the ability to learn from and work with graph-structured data, the project is expected to provide new tools for researchers across various scientific fields. The proposed research contributes to advancements in areas such as drug discovery, network analysis, and modeling of physical systems, offering new ways to approach complex problems in these domains. This project also offers research training opportunities for undergraduate and graduate students. The project focuses on four main research areas: (1) developing more expressive and efficient graph neural networks, (2) creating improved generative models for graphs, (3) applying graph learning techniques to optimization problems, and (4) exploring the use of graph neural networks for discovering physical relations. The interconnected research thrusts aim to improve the capabilities of machine learning models based on graphs, laying the groundwork for solving complex graph-related challenges. The project will produce new mathematical and statistical tools, theoretical frameworks, and assessment methods for learning from graphs. The work is expected to advance graph learning techniques and their applications in scientific fields, providing researchers with new ways to handle data structured as graphs. 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.