Poster (Scientific congresses and symposiums)
You say Normalizing Flows I see Bayesian Networks
Wehenkel, Antoine; Louppe, Gilles
2020ICML2020 Workshop on Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models
Peer reviewed
 

Files


Full Text
You_say_Normalizing_Flows_I_see_Bayesian_Networks.pdf
Author postprint (15.29 MB)
Download

All documents in ORBi are protected by a user license.

Send to



Details



Keywords :
Normalizing Flows; Bayesian Networks; Density Estimation
Abstract :
[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.
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 :
You say Normalizing Flows I see Bayesian Networks
Publication date :
10 July 2020
Number of pages :
Antoine Wehenkel
Event name :
ICML2020 Workshop on Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models
Event date :
July 2020
Audience :
International
Peer reviewed :
Peer reviewed
Name of the research project :
Deep learning for inverse problems
Funders :
F.R.S.-FNRS - Fonds de la Recherche Scientifique [BE]
Available on ORBi :
since 30 June 2020

Statistics


Number of views
66 (8 by ULiège)
Number of downloads
167 (7 by ULiège)

Bibliography


Similar publications



Contact ORBi