LANS Seminar

Seminar Title: Low-discrepancy designs via graph neural networks

Speaker: Nathan Kirk, Senior Research Associate in, The Center for Interdisciplinary Scientific Computation, Illinois Institute of Technology

Date/Time: Thursday, February 6, 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: Message-Passing Monte Carlo (MPMC) is the first machine-learning method for the generation of low-discrepancy (space-filling) point sets. We leverage tools from Geometric Deep Learning and base our model on Graph Neural Networks. MPMC achieves state-of-the-art performance with respect to discrepancy, a uniformity measure for space-filling designs, and is superior to previous methods by a significant margin. In fact, MPMC points are empirically shown to be either optimal or near-optimal with respect to the discrepancy in every instance where the optimal discrepancy can be determined. Lastly, we show a recent extension generating representative points from an arbitrary probability distribution given the density function.

Bio: Nathan Kirk is a Senior Research Associate in the Center for Interdisciplinary Scientific Computation at Illinois Institute of Technology. Before that, he was a Postdoctoral Scholar at University of Waterloo, Canada and got his PhD from Queen’s University Belfast, Northern Ireland. Nathan’s main research interests are Monte Carlo methods and variants and implementing state-of-the-art machine learning methods in an attempt to enhance and optimize sampling procedures.

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