Tag
Proposes SNAP-FM, a method that leverages sparse GPU nonlinear optimization to accelerate constraint projection in physics-constrained generative modeling, achieving faster inference while preserving exact physical constraint satisfaction.