Active Diffusion-Based Inference for Ill-Posed Inverse Problems Under Incomplete Priors

Inverse problems in the physical sciences are frequently ill-posed: many candidate solutions fit the observations equally well, and the prior knowledge that would ordinarily break the tie is itself incomplete. This work explores how diffusion-based methods can support active inference in exactly that setting — choosing what to measure next when the prior cannot be trusted to disambiguate on its own.

Recommended citation: Jitao Xu, N. Sato, Y. Li. (2026). "Active Diffusion-Based Inference for Ill-Posed Inverse Problems Under Incomplete Priors." Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence (IJCAI-ECAI 2026), Main Track.