Seminar abstract:
Many engineering settings involve large collections of physical assets where all assets are highly related to one another, yet also have certain key individual variations affecting their response to their environment and use. This presentation is concerned with developing data processing methodologies whereby one can efficiently leverage large amounts of response data from collections of physical systems to infer properties of the population of assets as a whole. We call this “distributional inversion.” We will look at examples of this methodology in problems of subsurface groundwater flow, structural dynamics, aerodynamics of wind turbine blades, and material mechanics. Furthermore, we show how we can accelerate the learning task by concurrently learning surrogate models which can efficiently approximate expensive numerical simulations.
Speaker bio:
Arnaud Vadeboncoeur will be joining Imperial College London in October 2026 as a Research Fellow (IRF) between the Department of Mathematics and Imperial’s I-X Center for AI in Science. He is currently a postdoc in the Department of Engineering at the University of Cambridge in the Computational Statistics and Machine Learning lab, working with Prof. Mark Girolami. Prior to this, he completed his PhD in Engineering at Cambridge with Prof. Fehmi Cirak and earned his undergraduate degree in Civil Engineering at the University of Ottawa.