AI RESEARCH
MCMC-Correction of Score-Based Diffusion Models for Model Composition
arXiv CS.LG
•
ArXi:2307.14012v4 Announce Type: replace-cross Diffusion models can be parameterized in terms of either score or energy function. The energy parameterization is attractive as it enables sampling procedures such as Marko Chain Monte Carlo (MCMC) that incorporates a Metropolis--Hastings (MH) correction step based on energy differences between proposed samples. Such corrections can significantly improve sampling quality, particularly in the context of model composition, where pre-trained models are combined to generate samples from novel distributions.