BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:The elephant in the room: fluid dynamics in the age of machine lea
 rning - Matthew Juniper\, CUED
DTSTART:20240119T160000Z
DTEND:20240119T170000Z
UID:TALK209845@talks.cam.ac.uk
CONTACT:Professor Grae Worster
DESCRIPTION:John von Neumann is often quoted as saying "with four paramete
 rs I can fit an elephant\, and with five I can make him wiggle his trunk."
  The implication seems to be that physical models should contain only a ha
 ndful of parameters. A century later\, however\, we are happy to use physi
 cs-agnostic neural networks containing millions of parameters. What would 
 von Neumann say? How should physical modellers respond?\n\nIn this talk\, 
 I will show that von Neumann's quote is more nuanced than it sounds. I wil
 l then frame a response within a Bayesian framework\, in which physical pr
 inciples such as conservation of mass and momentum are treated as high qua
 lity prior information\, with quantified uncertainty\, expressed as PDEs o
 r low order models. The information content of data can then be quantified
  and the likelihood of different candidate models can be compared after th
 e data arrives. I will show how Bayesian inference becomes computationally
  tractable when combined with adjoint methods. I will demonstrate this thr
 ough (i) assimilation of 3D Flow-MRI data into Finite Element CFD\, and (i
 i) assimilation of acoustics data into low order models\, which are then r
 anked by their likelihood given the data. The main message of the talk is 
 "keep the physics in the model if you can."\n
LOCATION:MR2
END:VEVENT
END:VCALENDAR