Diffusion Priors In Variational Autoencoders
This is an incremental improvement for generative modeling researchers, addressing the underperformance of VAEs compared to other models.
The authors tackled the problem of improving variational autoencoders (VAEs) by using denoising diffusion probabilistic models (DDPMs) as priors for latent variables, resulting in a model that is competitive with normalizing flow-based priors.
Among likelihood-based approaches for deep generative modelling, variational autoencoders (VAEs) offer scalable amortized posterior inference and fast sampling. However, VAEs are also more and more outperformed by competing models such as normalizing flows (NFs), deep-energy models, or the new denoising diffusion probabilistic models (DDPMs). In this preliminary work, we improve VAEs by demonstrating how DDPMs can be used for modelling the prior distribution of the latent variables. The diffusion prior model improves upon Gaussian priors of classical VAEs and is competitive with NF-based priors. Finally, we hypothesize that hierarchical VAEs could similarly benefit from the enhanced capacity of diffusion priors.