Prof Michael Shields, Johns Hopkins University
YouTube Stream: https://www.youtube.com/watch?v=B4P-hr-W7lw
Join group to receive calendar invite: https://groups.google.com/a/modelingtalks.org/g/talks
Abstract:
As Artificial Intelligence (AI) continues to transform the scientific enterprise, complex physical systems are increasingly being modeling with Scientific Machine Learning (SciML) and Scientific Foundation Models (SciFMs). Uncertainty Quantification (UQ) is an essential part of this modeling process for many reasons. These models are imperfect representations of real stochastic physical systems, trained from seemingly abundant but practically limited data sets, and used for high consequence decision making. UQ is critical for assessing and ensuring confidence in these model predictions. However, while SciML models and SciFMs have exploded in size, complexity, and predictive power over the past decade, UQ developments have not kept pace – impeded by theoretical and computational challenges alike. At the center of this bottleneck is the curse of dimensionality. In this work, I review several of the major advances in UQ for deep learning from the past decade with an emphasis on Bayesian methods, discussing their advantages and pitfalls in the context of modeling complex physical systems. I then offer a more comprehensive outlook on UQ for SciFMs that aims to account for the many different sources of uncertainty and present perspectives on achieving practical, rigorous, and scalable UQ methods for these systems.
Bio: Michael D. Shields is a Professor in the Dept. of Civil & Systems Engineering at Johns Hopkins University. Prof. Shields is the Director of the Center on High-Throughput Materials Discovery for Extremes (HT-MAX), Associate Director for AI in the AI for Materials Design Network (AIMD-Net), holds a secondary appointment in the Dept. of Materials Science and Engineering, is a fellow of the Hopkins Extreme Materials Institute and a member of the Data Science at AI Institute. Prof. Shields conducts research in uncertainty quantification (UQ) and machine learning for problems in mechanics, materials science, and physics with applications ranging from multi-scale material modeling to assessing the reliability and safety of large-scale structures. He received his Ph.D. in Civil Engineering and Engineering Mechanics from Columbia University in 2010. He then served as a Research Engineer in applied computational mechanics at Weidlinger Associates, Inc before he joined the faculty at Johns Hopkins in 2013. For his work in UQ, Prof. Shields has been awarded the ONR Young Investigator Award, the NSF CAREER Award, the DOE Early Career Award, and the IASSAR Early Achievement Award. Prof. Shields and his group also develop the open-source UQpy (Uncertainty Quantification with Python) software, which is a general toolbox and development environment for UQ in computational, mathematical, and physical systems.
Bio:
Michael D. Shields is a Professor in the Dept. of Civil & Systems Engineering at Johns Hopkins University. Prof. Shields is the Director of the Center on High-Throughput Materials Discovery for Extremes (HT-MAX), Associate Director for AI in the AI for Materials Design Network (AIMD-Net), holds a secondary appointment in the Dept. of Materials Science and Engineering, is a fellow of the Hopkins Extreme Materials Institute and a member of the Data Science at AI Institute. Prof. Shields conducts research in uncertainty quantification (UQ) and machine learning for problems in mechanics, materials science, and physics with applications ranging from multi-scale material modeling to assessing the reliability and safety of large-scale structures. He received his Ph.D. in Civil Engineering and Engineering Mechanics from Columbia University in 2010. He then served as a Research Engineer in applied computational mechanics at Weidlinger Associates, Inc before he joined the faculty at Johns Hopkins in 2013. For his work in UQ, Prof. Shields has been awarded the ONR Young Investigator Award, the NSF CAREER Award, the DOE Early Career Award, and the IASSAR Early Achievement Award. Prof. Shields and his group also develop the open-source UQpy (Uncertainty Quantification with Python) software, which is a general toolbox and development environment for UQ in computational, mathematical, and physical systems.