Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Aarts, Gert, Habibi, Diaa E., Wang, Lingxiao, Zhou, Kai
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2410.21212
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910903167877120
author Aarts, Gert
Habibi, Diaa E.
Wang, Lingxiao
Zhou, Kai
author_facet Aarts, Gert
Habibi, Diaa E.
Wang, Lingxiao
Zhou, Kai
contents To analyse how diffusion models learn correlations beyond Gaussian ones, we study the behaviour of higher-order cumulants, or connected n-point functions, under both the forward and backward process. We derive explicit expressions for the moment- and cumulant-generating functionals, in terms of the distribution of the initial data and properties of forward process. It is shown analytically that during the forward process higher-order cumulants are conserved in models without a drift, such as the variance-expanding scheme, and that therefore the endpoint of the forward process maintains nontrivial correlations. We demonstrate that since these correlations are encoded in the score function, higher-order cumulants are learnt in the backward process, also when starting from a normal prior. We confirm our analytical results in an exactly solvable toy model with nonzero cumulants and in scalar lattice field theory.
format Preprint
id arxiv_https___arxiv_org_abs_2410_21212
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle On learning higher-order cumulants in diffusion models
Aarts, Gert
Habibi, Diaa E.
Wang, Lingxiao
Zhou, Kai
High Energy Physics - Lattice
Disordered Systems and Neural Networks
Machine Learning
To analyse how diffusion models learn correlations beyond Gaussian ones, we study the behaviour of higher-order cumulants, or connected n-point functions, under both the forward and backward process. We derive explicit expressions for the moment- and cumulant-generating functionals, in terms of the distribution of the initial data and properties of forward process. It is shown analytically that during the forward process higher-order cumulants are conserved in models without a drift, such as the variance-expanding scheme, and that therefore the endpoint of the forward process maintains nontrivial correlations. We demonstrate that since these correlations are encoded in the score function, higher-order cumulants are learnt in the backward process, also when starting from a normal prior. We confirm our analytical results in an exactly solvable toy model with nonzero cumulants and in scalar lattice field theory.
title On learning higher-order cumulants in diffusion models
topic High Energy Physics - Lattice
Disordered Systems and Neural Networks
Machine Learning
url https://arxiv.org/abs/2410.21212