Reference : You say Normalizing Flows I see Bayesian Networks
Scientific congresses and symposiums : Unpublished conference/Abstract
Engineering, computing & technology : Computer science
http://hdl.handle.net/2268/249062
You say Normalizing Flows I see Bayesian Networks
English
Wehenkel, Antoine mailto [Université de Liège - ULiège > Dép. d'électric., électron. et informat. (Inst.Montefiore) > Big Data >]
Louppe, Gilles mailto [Université de Liège - ULiège > Dép. d'électric., électron. et informat. (Inst.Montefiore) > Big Data >]
10-Jul-2020
Antoine Wehenkel
ICML2020 Workshop on Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models
Yes
No
International
ICML2020 Workshop on Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models
July 2020
[en] Normalizing Flows ; Bayesian Networks ; Density Estimation
[en] Normalizing flows have emerged as an important family of deep neural networks for modelling complex probability distributions. In this note, we
revisit their coupling and autoregressive transformation layers as probabilistic graphical models and show that they reduce to Bayesian networks with a pre-defined topology and a learnable density at each node. From this new perspective, we provide three results. First, we show that stacking multiple transformations in a normalizing flow relaxes independence assumptions and entangles the model distribution. Second, we show that a fundamental leap of capacity emerges when the depth of affine flows exceeds 3 transformation layers. Third, we prove the non-universality of the affine normalizing flow, regardless of its depth.
F.R.S.-FNRS - Fonds de la Recherche Scientifique
Deep learning for inverse problems
Researchers ; Professionals ; Students
http://hdl.handle.net/2268/249062

File(s) associated to this reference

Fulltext file(s):

FileCommentaryVersionSizeAccess
Open access
You_say_Normalizing_Flows_I_see_Bayesian_Networks.pdfAuthor postprint14.93 MBView/Open

Bookmark and Share SFX Query

All documents in ORBi are protected by a user license.