Likelihood Matching for Diffusion Models
This addresses training efficiency and theoretical guarantees for diffusion models, which are incremental improvements to existing diffusion model frameworks.
The authors tackled the problem of training diffusion models by proposing a Likelihood Matching approach that establishes equivalence between target data likelihood and reverse diffusion sample path likelihood, using a quasi-likelihood approximation with Gaussian distributions matched on conditional mean and covariance. They demonstrated effectiveness through empirical evaluations and provided non-asymptotic convergence guarantees for their sampler.
We propose a Likelihood Matching approach for training diffusion models by first establishing an equivalence between the likelihood of the target data distribution and a likelihood along the sample path of the reverse diffusion. To efficiently compute the reverse sample likelihood, a quasi-likelihood is considered to approximate each reverse transition density by a Gaussian distribution with matched conditional mean and covariance, respectively. The score and Hessian functions for the diffusion generation are estimated by maximizing the quasi-likelihood, ensuring a consistent matching of both the first two transitional moments between every two time points. A stochastic sampler is introduced to facilitate computation that leverages on both the estimated score and Hessian information. We establish consistency of the quasi-maximum likelihood estimation, and provide non-asymptotic convergence guarantees for the proposed sampler, quantifying the rates of the approximation errors due to the score and Hessian estimation, dimensionality, and the number of diffusion steps. Empirical and simulation evaluations demonstrate the effectiveness of the proposed Likelihood Matching and validate the theoretical results.