<ul><li>Diffusion models have become popular for modeling physical systems governed by partial differential equations (PDEs).</li><li>Constraints on clean samples are typically applied to the expectation of clean samples, leading to a trade-off in generative modeling accuracy known as Jensen's Gap.</li><li>A new approach called Physics-Informed Distillation of Diffusion Models (PIDDM) proposes post-hoc distillation to enforce PDE constraints effectively.</li><li>PIDDM improves PDE satisfaction in generative modeling while supporting forward and inverse problem solving, with less computation overhead compared to other baselines.</li></ul>

Physics-Informed Distillation of Diffusion Models for PDE-Constrained Generation

Discover more