AI与世界难题 · 物理学 · arXiv · 2026-10-20
Data Completion for Electrical Impedance Tomography by Conditional Diffusion Models
Data scarcity is a fundamental barrier in Electrical Impedance Tomography (EIT), as undersampled Dirichlet-to-Neumann (DtN) measurements can substantially degrade conductivity reconstructions. The authors train a conditional diffusion model to learn the distribution of DtN data and infer full measurement vectors from partial observations, and the completion step can be used as a plug-in preprocessing stage with off-the-shelf EIT solvers. In numerical experiments, diffusion completion achieved reconstructions comparable to the full-data baseline using only 1% of the measurements, whereas standard baselines such as matrix completion required 30% to reach similar quality.