Professor Matthew Juniper - The Elephant in the Room: Fluid Dynamics in the age of Machine Learning Abstract"With four parameters I can fit an elephant, and with five I can make him wiggle his trunk," said John von Neumann in a powerful exhortation that physical models should contain only a handful of parameters. A century later, we seem happy to use physics-agnostic neural networks containing millions of parameters. What would von Neumann say? How should physical modellers respond?In this talk, I will frame a response within a Bayesian framework, in which physical principles such as conservation of mass and momentum are expressed as high quality prior information with quantified uncertainties. I will show how Bayesian inference becomes computationally tractable when combined with adjoint methods, and demonstrate this through assimilation of 3D Flow-MRI data directly into CFD, and selection of models in an acoustic problem.BiographyMatthew Juniper is Professor of Thermofluid Mechanics at the University of Cambridge. His research interests are flow instability, adjoint-based sensitivity analysis, shape optimization, and physics-based Bayesian inference, particularly when accelerated with adjoint methods. He is an Associate Editor of the Journal of Fluid Mechanics and PI of the UK Fluids Network. Dec 07 2023 15.00 - 16.00 Professor Matthew Juniper - The Elephant in the Room: Fluid Dynamics in the age of Machine Learning The Edinburgh Fluid Dynamics Group is pleased to welcome Professor Matthew Juniper as the next speaker in our new series of colloquia. This colloquium continues an exciting new series co-funded by the Schools of Engineering, Mathematics, and Physics & Astronomy. The series will regularly invite high-profile speakers to give talks aimed at a broad audience of researchers interested in fluid dynamics. Elm Lecture Theatre, The Nucleus, Kings Buildings This article was published on Wednesday 2 September 2026
Professor Matthew Juniper - The Elephant in the Room: Fluid Dynamics in the age of Machine Learning Abstract"With four parameters I can fit an elephant, and with five I can make him wiggle his trunk," said John von Neumann in a powerful exhortation that physical models should contain only a handful of parameters. A century later, we seem happy to use physics-agnostic neural networks containing millions of parameters. What would von Neumann say? How should physical modellers respond?In this talk, I will frame a response within a Bayesian framework, in which physical principles such as conservation of mass and momentum are expressed as high quality prior information with quantified uncertainties. I will show how Bayesian inference becomes computationally tractable when combined with adjoint methods, and demonstrate this through assimilation of 3D Flow-MRI data directly into CFD, and selection of models in an acoustic problem.BiographyMatthew Juniper is Professor of Thermofluid Mechanics at the University of Cambridge. His research interests are flow instability, adjoint-based sensitivity analysis, shape optimization, and physics-based Bayesian inference, particularly when accelerated with adjoint methods. He is an Associate Editor of the Journal of Fluid Mechanics and PI of the UK Fluids Network. Dec 07 2023 15.00 - 16.00 Professor Matthew Juniper - The Elephant in the Room: Fluid Dynamics in the age of Machine Learning The Edinburgh Fluid Dynamics Group is pleased to welcome Professor Matthew Juniper as the next speaker in our new series of colloquia. This colloquium continues an exciting new series co-funded by the Schools of Engineering, Mathematics, and Physics & Astronomy. The series will regularly invite high-profile speakers to give talks aimed at a broad audience of researchers interested in fluid dynamics. Elm Lecture Theatre, The Nucleus, Kings Buildings This article was published on Wednesday 2 September 2026
Dec 07 2023 15.00 - 16.00 Professor Matthew Juniper - The Elephant in the Room: Fluid Dynamics in the age of Machine Learning The Edinburgh Fluid Dynamics Group is pleased to welcome Professor Matthew Juniper as the next speaker in our new series of colloquia. This colloquium continues an exciting new series co-funded by the Schools of Engineering, Mathematics, and Physics & Astronomy. The series will regularly invite high-profile speakers to give talks aimed at a broad audience of researchers interested in fluid dynamics.
