Polygonal Unadjusted Langevin Algorithms: Creating stable and efficient adaptive algorithms for neural networks

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lim, Dong-Young, Sabanis, Sotirios
Format: Preprint
Published: 2021
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911787333451776
author Lim, Dong-Young
Sabanis, Sotirios
author_facet Lim, Dong-Young
Sabanis, Sotirios
contents We present a new class of Langevin based algorithms, which overcomes many of the known shortcomings of popular adaptive optimizers that are currently used for the fine tuning of deep learning models. Its underpinning theory relies on recent advances of Euler's polygonal approximations for stochastic differential equations (SDEs) with monotone coefficients. As a result, it inherits the stability properties of tamed algorithms, while it addresses other known issues, e.g. vanishing gradients in neural networks. In particular, we provide a nonasymptotic analysis and full theoretical guarantees for the convergence properties of an algorithm of this novel class, which we named TH$\varepsilon$O POULA (or, simply, TheoPouLa). Finally, several experiments are presented with different types of deep learning models, which show the superior performance of TheoPouLa over many popular adaptive optimization algorithms.
format Preprint
id arxiv_https___arxiv_org_abs_2105_13937
institution arXiv
publishDate 2021
record_format arxiv
spellingShingle Polygonal Unadjusted Langevin Algorithms: Creating stable and efficient adaptive algorithms for neural networks
Lim, Dong-Young
Sabanis, Sotirios
Machine Learning
Optimization and Control
Probability
We present a new class of Langevin based algorithms, which overcomes many of the known shortcomings of popular adaptive optimizers that are currently used for the fine tuning of deep learning models. Its underpinning theory relies on recent advances of Euler's polygonal approximations for stochastic differential equations (SDEs) with monotone coefficients. As a result, it inherits the stability properties of tamed algorithms, while it addresses other known issues, e.g. vanishing gradients in neural networks. In particular, we provide a nonasymptotic analysis and full theoretical guarantees for the convergence properties of an algorithm of this novel class, which we named TH$\varepsilon$O POULA (or, simply, TheoPouLa). Finally, several experiments are presented with different types of deep learning models, which show the superior performance of TheoPouLa over many popular adaptive optimization algorithms.
title Polygonal Unadjusted Langevin Algorithms: Creating stable and efficient adaptive algorithms for neural networks
topic Machine Learning
Optimization and Control
Probability
url https://arxiv.org/abs/2105.13937