Tuesday, October 13, 2026
4:00 PM – 5:00 PM CDT
Schmidt AI in Science Speaker Series: Doug Brinkerhoff (University of Montana)
About this event
A free talk at UChicago on why a machine learning glacier flow emulator ended up back as a PDE solver, covering GPU-accelerated multigrid methods and ML-inspired transport operators in scientific computing.
Doug Brinkerhoff of the University of Montana presents to the Schmidt AI in Science Speaker Series at the University of Chicago.
Talk: "Why my glacier flow emulator turned back into a PDE solver: GPU-accelerated multigrid, ML-inspired transport operators, and the difficulties that necessitated them."
This is a genuinely useful counterweight to the usual surrogate-model optimism. The talk traces a project that set out to replace a physics solver with a learned emulator and instead circled back to solving the partial differential equations directly, with the machine learning influence surviving in the design of the transport operators and the GPU-accelerated multigrid implementation rather than in the model itself.
Relevant to anyone working on scientific machine learning, physics-informed models, or surrogate modeling who wants an honest account of where learned emulators break down and what you keep from the attempt.
Format. Light refreshments from 3:45 pm, talk from 4:00 to 4:45 pm, Q&A from 4:45 to 5:00 pm, followed by a reception.
Free and open, in Room 390 of the John Crerar Library.
Organized by
The University of Chicago's hub for data science and AI research, education and partnerships. The institute runs public speaker series, conferences and fellowship programs spanning AI methods and their application across scientific disciplines.
All events by University of Chicago Data Science Institute →