Neural Flow Diffusion Models: Learnable Forward Process for Improved Diffusion Modelling

arXiv:2404.12940v1 Announce Type: cross
Abstract: Conventional diffusion models typically relies on a fixed forward process, which implicitly defines complex marginal distributions over latent variables. This can often complicate the reverse process' task in learning generative trajectories, and results in costly inference for diffusion models. To address these limitations, we introduce Neural Flow Diffusion Models (NFDM), a novel framework that enhances diffusion models by supporting a broader range of forward processes beyond the fixed linear Gaussian. We also propose a novel …

abstract arxiv cs.cv cs.lg diffusion diffusion models flow generative inference limitations modelling process results stat.ml type variables