LANS Seminar

Seminar Title: Algorithms for GNN Explanations and GNNs for Learning Combinatorial Algorithms

Speaker: Sourav Medya, Assistant Professor, Department of Computer Science, University of Illinois Chicago (UIC)

Date/Time: Thursday, April 17, 2025/ 2:30 PM – 3:30 PM (Virual)
Location: See Meeting URL on the cels-seminars website which will require an Argonne login.

Description: Graph neural networks (GNNs) are powerful graph-based machine-learning models that are popular in various domains, e.g., social media, transportation, and drug discovery. However, owing to complex data representations, GNNs do not easily allow for human-intelligible explanations of their predictions. In this talk, I will present two complementary lines of research that advance our understanding of GNNs from an algorithmic perspective. First, I will introduce GraphTrail, a global post-hoc explainer that uncovers the combinatorial reasoning learned by GNNs. Unlike traditional instance-level explainers, GraphTrail automatically mines subgraph-level concepts and uses symbolic regression to express a model’s predictions as logical formulas and offers faithful and human-interpretable explanations. Second, I will present NeuroCUT, a novel GNN-based framework for solving NP-hard graph partitioning problems. NeuroCUT is a generic framework based on reinforcement learning to handle different non-differentiable objectives and supports arbitrary partition counts at inference time. Thus, it goes beyond traditional graph algorithms and enables inductive generalization across diverse combinatorial settings. The last part of the talk will demonstrate a few exciting future directions involving different application scenarios.

Bio: Sourav Medya is an Assistant Professor at the Department of Computer Science, University of Illinois Chicago (UIC). Before joining UIC, he was a research assistant professor in the Kellogg School of Management at Northwestern University and the Northwestern Institute on Complex Systems (NICO). He received his Ph.D. in Computer Science from University of California, Santa Barbara, and he got his Master of Engineering degree in Computer Science and Automation from Indian Institute of Science (IISc), Bangalore, India. His research focuses on the problems at the intersection of graphs, machine learning, and data science with a focus on explainability. He also build machine learning based techniques that have high impact in the areas such as healthcare, infrastructure, and computational social science.

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