Poster (Scientific congresses and symposiums)
Diffusion Priors In Variational Autoencoders
Wehenkel, Antoine; Louppe, Gilles
2021ICML Workshop on Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models
Peer reviewed
 

Files


Full Text
diffusion_priors_in_variationa.pdf
Publisher postprint (1.33 MB)
Download

All documents in ORBi are protected by a user license.

Send to



Details



Keywords :
Diffusion models; Generative models; Variational auto-encoders; Normalizing flows; DDPM; VAE
Abstract :
[en] 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.
Disciplines :
Computer science
Author, co-author :
Wehenkel, Antoine  ;  Université de Liège - ULiège > Dép. d'électric., électron. et informat. (Inst.Montefiore) > Big Data
Louppe, Gilles  ;  Université de Liège - ULiège > Dép. d'électric., électron. et informat. (Inst.Montefiore) > Big Data
Language :
English
Title :
Diffusion Priors In Variational Autoencoders
Alternative titles :
[en] Diffusion Priors In Variational Autoencoders
Publication date :
July 2021
Number of pages :
6
Event name :
ICML Workshop on Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models
Event organizer :
Chin-Wei Huang, David Krueger, Rianne van den Berg, George Papamakarios, Ricky Chen, Danilo Rezende
Event date :
July 2021
Audience :
International
Peer reviewed :
Peer reviewed
Funders :
F.R.S.-FNRS - Fonds de la Recherche Scientifique
Available on ORBi :
since 29 July 2021

Statistics


Number of views
215 (8 by ULiège)
Number of downloads
196 (4 by ULiège)

Bibliography


Similar publications



Contact ORBi