Graph AI: Science and Industrial Applications

Industrial Problems Seminar 

Jie Chen (IBM Research)

Abstract

Graphs serve as both a mathematical abstraction and a structured framework for organizing data, finding widespread applications across scientific and technological domains. The ascent of graph neural networks underscores their exceptional efficacy in capturing intricate data interactions, leading to a resurgence of traditional applications with elevated solution quality and the emergence of novel uses. This talk delves into several graph-related challenges encountered in industrial contexts and the consequent evolution of graph-based deep learning methodologies. Topics include the learning of graph grammar for advancing material discovery and circuit design, the scaling of graph neural network training for financial forensics, and the unveiling of latent graph structures in power grid analytics. The talk concludes with a discussion on graph-based learning in the era of foundation models and research opportunities.