{"id":"arxiv/2604.11375","url":"https://arxiv.org/abs/2604.11375","tags":["math.NA"],"score":0,"title":"DiLO: Decoupling Generative Priors and Neural Operators via Diffusion Latent Optimization for Inverse Problems","domain":"math.NA","source":"arxiv","authors":["Haibo Liu","Guang Lin"],"reviews":0,"subject":"math.NA","summary":"Diffusion models have emerged as powerful generative priors for solving PDE-constrained inverse problems. Compared with end-to-end approaches that learn inverse mappings from large paired observation--parameter datasets, decoupling the parameter prior from the forward physical model reduces reliance on paired data and improves flexibility. Fast neural operators can accelerate inference in such decoupled frameworks, but standard diffusion guidance evaluates them on partially denoised, non-physical intermediate states, introducing out-of-distribution (OOD) inputs. We therefore formulate the…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-04-13","reactions":{}}