Full Professor, University of California, San Diego
3 papers at NeurIPS 2025
We introduce neural approximation frameworks for approximating a family of geometric shape-fitting problems.
We introduce the Birkhoff Extension, a polynomial-time, almost-everywhere differentiable relaxation of permutation objectives to doubly stochastic matrices, enabling efficient combinatorial optimization.
Diffusion noise seeds contain universal compositional blueprints that persist across datasets, enabling our Patch PCA framework for zero-shot structure control.
