Don't Throw Out the Textbook: Old-School Graphics for New-School AI
Tiberiu Popa - Concordia University
Oct. 30, 2026, 3:30 p.m. - Oct. 30, 2026, 4:30 p.m.
ENGMC 11
Hosted by: Paul Kry
Today's AI can turn a sentence into a 3D dinosaur. So why bother learning the old material, like meshes, deformation energies, sampling theory and physics? Isn't all that about to be replaced by one big neural network?
Generative models, neural fields and Gaussian Splatting have changed how we build 3D content. We can now reconstruct scenes from photos, generate objects from text, and train networks to stand in for expensive simulations. So, is the classic graphics textbook obsolete? Learning-based methods are powerful but often lack fine geometric control, get noisy gradients from 2D supervision, waste capacity on uniform parameterizations, and run too slowly for interactive use. We'll look at a few examples where the most effective fixes came from classic graphics ideas built directly into the learning pipeline.
The take-home message applies to any field: the "old stuff" isn't obsolete. So, hang on to that textbook!!!
Tiberiu Popa is an Associate Professor at Concordia University. His main research interests are geometric modeling and optimization, geometric acquisition, performance capture, physics simulation, machine learning. Prior to joining Concordia in 2013, he was a Senior Researcher at ETH Zurich and earned a PhD from the University of British Columbia.
