IPAB Guest Speaker - 16/11/2026

Abstract: Many problems in geometry processing boil down to optimizing over fields (direction fields, phase fields, etc.) with singularities. In this setting, common optimization energies can exhibit pathological local optima or discretization-dependence, making for unreliable algorithms. I will show how these difficulties result from topology and how lifting to higher dimension can alleviate them. First, I will discuss how optimizing over graphs of fields gives us a nice way to encode singularities directly. Second, I will show how semidefinite relaxation amounts to low-dimensional lifting, yielding families of singularity configurations.

Bio: David is an applied geometry researcher who works on problems at the intersection of geometry, optimization, and physics, especially those of a topological character. He is currently a postdoc in the Visual Computing group at ISTA with Chris Wojtan. He did his PhD at MIT with Justin Solomon and spent two years as a postdoc in a physics group at Harvard with L. Mahadevan. Prior to that, he studied math at Cambridge and computer science at Harvard and interned at Pixar Research, where he helped build geometry and physics simulation tools for artists.