Abstract
Gaussian denoising has emerged as a powerful method for constructingsimulation-free continuous normalizing flows for generative modeling. Despitetheir empirical successes, theoretical properties of these flows and theregularizing effect of Gaussian denoising have remained largely unexplored. Inthis work, we aim to address this gap by investigating the well-posedness ofsimulation-free continuous normalizing flows built on Gaussian denoising.Through a unified framework termed Gaussian interpolation flow, we establishthe Lipschitz regularity of the flow velocity field, the existence anduniqueness of the flow, and the Lipschitz continuity of the flow map and thetime-reversed flow map for several rich classes of target distributions. Thisanalysis also sheds light on the auto-encoding and cycle consistency propertiesof Gaussian interpolation flows. Additionally, we study the stability of theseflows in source distributions and perturbations of the velocity field, usingthe quadratic Wasserstein distance as a metric. Our findings offer valuableinsights into the learning techniques employed in Gaussian interpolation flowsfor generative modeling, providing a solid theoretical foundation forend-to-end error analyses of learning Gaussian interpolation flows withempirical observations.