LANS Seminar

Seminar Title: Graph Learning for Robustness and Anomaly Detection in Complex Networks

Speaker: Hongwei Jin, Postdoctoral Researcher, Mathematics and Computer Science Division, Argonne National Laboratory

Date/Time: Thursday, January 16, 2025/ 2:30 PM – 3:30 PM (In-Person)
Location: See Meeting URL on the cels-seminars website which will require an Argonne login.

Description: Graph-based machine learning has become an essential method for modeling intricate data structures, yet it faces significant challenges, including susceptibility to adversarial attacks and the curse of dimensionality. This talk will present an innovative framework designed to certify the robustness of graph neural networks (GNNs) concerning permutation invariance, which offers theoretical assurances of resilience against topological perturbations, facilitating the use of GNNs in safety-critical applications. The presentation will then address the anomaly detection problem within computational workflows, focusing on identifying abnormal execution patterns that diverge from standard behavior. This involves harnessing the strengths of GNNs through both supervised and unsupervised approaches. Following this, exciting research directions that emerged from this work, including the development of graph foundation models, self-supervised learning frameworks, and their applications to novel domains will be briefly discussed.

Bio: Hongwei Jin is a postdoctoral researcher at the Mathematics and Computer Science division of Argonne National Laboratory. He earned his Ph.D. in Computer Science from the University of Illinois at Chicago in 2022, following an M.S. in Applied Mathematics from the Illinois Institute of Technology. His research focuses on developing and applying advanced machine learning techniques, including graph learning, geometric deep learning, and optimization methods, to tackle critical challenges in scientific domains. His current work investigates the resilience of scientific workflows by leveraging the power of large language models and incorporating principles of swarm intelligence, as well as a broader interest in AI surrogates for scientific tasks such as power grid modeling and analysis.

Please note that the meeting URL for this event can be seen on the cels-seminars website which requires an Argonne login.