The paper presents a novel approach to generative models, extending the continuous time framework from Brownian motion to fractional Brownian motion (FBM). The authors propose Generative Fractional Diffusion Models (GFDM), which utilize a reparameterization trick and a reverse time model. The process is driven by noise that converges to a non-Markovian process of infinite quadratic variation. The Hurst index H (0,1) of FBM allows for control of the distribution transforming path. The paper represents the first attempt to build a generative model based on a stochastic process with infinite quadratic variation.
Publication date: 26 Oct 2023
Project Page: https://arxiv.org/abs/2310.17638v1
Paper: https://arxiv.org/pdf/2310.17638
