Name: Ravi Patel
Pronouns: he/him/his
Biography:
Ravi is a researcher in the scientific machine learning department at Sandia National Labs. He primarily works on methods of extracting surrogate models of physical systems using high fidelity simulations. He has previously developed the operator learning methods, MOR-Physics and GMLS-Nets. He also works on incorporating physical constraints into machine learned models and uncertainty quantification for neural networks.
Institution/Lab: Sandia National Laboratories
Website: https://scholar.google.com/citations?user=BB14UP4AAAAJ&hl=en&oi=sra
SRP Collaboration Topic/Title: Uncertainty quantification for operator learning under physics informed constraints
Field or research area: Scientifc Machine Learning
Please select all the topical areas that apply to your project:
Computational Science Applications (i.e., bioscience, cosmology, chemistry, environmental science, nanotechnology, climate, etc.); Data Science (i.e., data analytics, data management & storage systems, visualization); Machine Learning and AI
Brief Abstract:
One promising strategy in scientific machine learning seeks to obtain a PDE description of a physical system by inferring the PDE’s operators directly from observations of the system. Ideally, a modeler would also enforce a priori known physics constraints such as conservation and symmetries to guarantee the model is at least physically valid. However, observational data is always noisy and sparse, so learned operators are only useful insofar as their uncertainties have been properly quantified. Without these measures, an analyst using the models cannot determine in which regimes the model is valid. The staff members have previously developed methods for learning uncertainty aware operators. The intern will focus on combining these operators with physics informed constraints and studying the interaction. While the methods developed will be broadly applicable, we will focus on two exemplars, crack propagation of a composite under shear loading and climate forcing under volcanic eruption. This topic will blend together ideas from a variety of subjects, e.g., PDEs, statistics, machine learning, solid mechanics, and climatology.
Desired relevant skills, background, or interests:
(Not necessarily required) familiarity with coding, machine learning, Bayesian inference, PDEs, and HPC
Other comments:
Do any special requirements apply? Permanent Resident OK; International OK
Other, specify:
Keywords:
Operator Learning; uncertainty quantification; physics-informed constraints; surrogate modelling
Lightning Talk Title: Operator Learning – an Ingredient for PDE Modelling