CoolMomentum: a method for stochastic optimization by Langevin dynamics with simulated annealing

Abstract Deep learning applications require global optimization of non-convex objective functions, which have multiple local minima. The same problem is often found in physical simulations and may be resolved by the methods of Langevin dynamics with Simulated Annealing, which is a well-established a...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Oleksandr Borysenko, Maksym Byshkin
Formato:	article
Lenguaje:	EN
Publicado:	Nature Portfolio 2021
Materias:	Medicine R Science Q
Acceso en línea:	https://doaj.org/article/9ae814904dc54f65a9749af760cb5d73
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:9ae814904dc54f65a9749af760cb5d73
record_format	dspace
spelling	oai:doaj.org-article:9ae814904dc54f65a9749af760cb5d732021-12-02T15:52:55ZCoolMomentum: a method for stochastic optimization by Langevin dynamics with simulated annealing10.1038/s41598-021-90144-32045-2322https://doaj.org/article/9ae814904dc54f65a9749af760cb5d732021-05-01T00:00:00Zhttps://doi.org/10.1038/s41598-021-90144-3https://doaj.org/toc/2045-2322Abstract Deep learning applications require global optimization of non-convex objective functions, which have multiple local minima. The same problem is often found in physical simulations and may be resolved by the methods of Langevin dynamics with Simulated Annealing, which is a well-established approach for minimization of many-particle potentials. This analogy provides useful insights for non-convex stochastic optimization in machine learning. Here we find that integration of the discretized Langevin equation gives a coordinate updating rule equivalent to the famous Momentum optimization algorithm. As a main result, we show that a gradual decrease of the momentum coefficient from the initial value close to unity until zero is equivalent to application of Simulated Annealing or slow cooling, in physical terms. Making use of this novel approach, we propose CoolMomentum—a new stochastic optimization method. Applying Coolmomentum to optimization of Resnet-20 on Cifar-10 dataset and Efficientnet-B0 on Imagenet, we demonstrate that it is able to achieve high accuracies.Oleksandr BorysenkoMaksym ByshkinNature PortfolioarticleMedicineRScienceQENScientific Reports, Vol 11, Iss 1, Pp 1-8 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	Medicine R Science Q
spellingShingle	Medicine R Science Q Oleksandr Borysenko Maksym Byshkin CoolMomentum: a method for stochastic optimization by Langevin dynamics with simulated annealing
description	Abstract Deep learning applications require global optimization of non-convex objective functions, which have multiple local minima. The same problem is often found in physical simulations and may be resolved by the methods of Langevin dynamics with Simulated Annealing, which is a well-established approach for minimization of many-particle potentials. This analogy provides useful insights for non-convex stochastic optimization in machine learning. Here we find that integration of the discretized Langevin equation gives a coordinate updating rule equivalent to the famous Momentum optimization algorithm. As a main result, we show that a gradual decrease of the momentum coefficient from the initial value close to unity until zero is equivalent to application of Simulated Annealing or slow cooling, in physical terms. Making use of this novel approach, we propose CoolMomentum—a new stochastic optimization method. Applying Coolmomentum to optimization of Resnet-20 on Cifar-10 dataset and Efficientnet-B0 on Imagenet, we demonstrate that it is able to achieve high accuracies.
format	article
author	Oleksandr Borysenko Maksym Byshkin
author_facet	Oleksandr Borysenko Maksym Byshkin
author_sort	Oleksandr Borysenko
title	CoolMomentum: a method for stochastic optimization by Langevin dynamics with simulated annealing
title_short	CoolMomentum: a method for stochastic optimization by Langevin dynamics with simulated annealing
title_full	CoolMomentum: a method for stochastic optimization by Langevin dynamics with simulated annealing
title_fullStr	CoolMomentum: a method for stochastic optimization by Langevin dynamics with simulated annealing
title_full_unstemmed	CoolMomentum: a method for stochastic optimization by Langevin dynamics with simulated annealing
title_sort	coolmomentum: a method for stochastic optimization by langevin dynamics with simulated annealing
publisher	Nature Portfolio
publishDate	2021
url	https://doaj.org/article/9ae814904dc54f65a9749af760cb5d73
work_keys_str_mv	AT oleksandrborysenko coolmomentumamethodforstochasticoptimizationbylangevindynamicswithsimulatedannealing AT maksymbyshkin coolmomentumamethodforstochasticoptimizationbylangevindynamicswithsimulatedannealing
_version_	1718385586923896832

CoolMomentum: a method for stochastic optimization by Langevin dynamics with simulated annealing

Ejemplares similares