Stochastic Backpropagation through Fourier Transforms - IMT Atlantique Accéder directement au contenu
Communication Dans Un Congrès Année : 2021

Stochastic Backpropagation through Fourier Transforms

Résumé

Backpropagating gradients through random variables is at the heart of numerous machine learning applications. In this paper, we present a general framework for deriving stochastic backpropagation rules for any continuous distribution. Our approach exploits the link between the characteristic function and the Fourier transform, to transport the derivatives from the parameters of the distribution to the random variable. Our method generalizes previously known estimators, and results in new estimators for the gamma, beta, Dirichlet and Laplace distributions. Furthermore, we show that the classical deterministic backproapagation rule in neural networks, is a special case of stochastic backpropagation with Dirac distributions, thus providing a link between probabilistic graphical models and neural networks.
Fichier principal
Vignette du fichier
paper_1048_cameraready.pdf (2.11 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04165289 , version 1 (07-08-2023)

Identifiants

Citer

Amine Echraibi, Joachim Flocon Cholet, Stéphane Gosselin, Sandrine Vaton. Stochastic Backpropagation through Fourier Transforms. 29th European Signal Processing Conference (EUSIPCO), Aug 2021, Dublin ( virtual ), Ireland. ⟨10.23919/EUSIPCO54536.2021.9616294⟩. ⟨hal-04165289⟩
19 Consultations
16 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More