Abstract
Many inverse problems are ill-posed and over-parameterized, making classical numerical inversion methods unstable and computationally expensive. While machine learning frameworks have shown promise in addressing these challenges, they risk overfitting to specific features of the training data if not informed by the underlying physics—often producing physically inconsistent results. This talk discusses recent efforts in physics-guided machine learning, with applications in acoustic wavefield inversion, nondestructive evaluation (NDE), and hyperspectral imaging (HSI). The presentation begins with a theoretical acoustic inverse problem to demonstrate how physics-informed training objectives support the learning of consistent inverse mappings. Applications in acoustic NDE follow, highlighting the role of domain geometry and governing physical operators in learning tasks such as estimating local thickness or detecting cracks based on the propagation characteristics of guided acoustic waves. Finally, the discussion extends to remote sensing, showcasing a physics-aware variational model for HSI measurement inversion.
About the Speaker
Josh Tempelman is a postdoctoral fellow in the Space Remote Sensing and Data Science group at Los Alamos National Laboratory. His current research focuses on applying scientific machine learning to inverse modeling and remote sensing, with particular emphasis on acoustical and hyperspectral phenomenologies. He earned his Ph.D. in mechanical engineering from the University of Illinois Urbana-Champaign in 2024, where he conducted research on mechanisms for passive energy management in acoustical and dynamical systems. He holds a B.S. in mechanical engineering from Calvin College and an M.S. from Michigan State University. He is the recipient of the National Science Foundation Graduate Research Fellowship (2020), the Los Alamos Distinguished Student award (2022) and Director’s Funded Postdoctoral Fellowship (2024), and the Stanley I. Weiss Outstanding Thesis in Mechanical Engineering (2025).
Host: Professor Alex Vakakis