Cargando…

Sampling can be faster than optimization

Optimization algorithms and Monte Carlo sampling algorithms have provided the computational foundations for the rapid growth in applications of statistical machine learning in recent years. There is, however, limited theoretical understanding of the relationships between these 2 kinds of methodology...

Descripción completa

Detalles Bibliográficos
Autores principales: Ma, Yi-An, Chen, Yuansi, Jin, Chi, Flammarion, Nicolas, Jordan, Michael I.
Formato: Online Artículo Texto
Lenguaje:English
Publicado: National Academy of Sciences 2019
Materias:
Acceso en línea:https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6800351/
https://www.ncbi.nlm.nih.gov/pubmed/31570618
http://dx.doi.org/10.1073/pnas.1820003116
_version_ 1783460428900204544
author Ma, Yi-An
Chen, Yuansi
Jin, Chi
Flammarion, Nicolas
Jordan, Michael I.
author_facet Ma, Yi-An
Chen, Yuansi
Jin, Chi
Flammarion, Nicolas
Jordan, Michael I.
author_sort Ma, Yi-An
collection PubMed
description Optimization algorithms and Monte Carlo sampling algorithms have provided the computational foundations for the rapid growth in applications of statistical machine learning in recent years. There is, however, limited theoretical understanding of the relationships between these 2 kinds of methodology, and limited understanding of relative strengths and weaknesses. Moreover, existing results have been obtained primarily in the setting of convex functions (for optimization) and log-concave functions (for sampling). In this setting, where local properties determine global properties, optimization algorithms are unsurprisingly more efficient computationally than sampling algorithms. We instead examine a class of nonconvex objective functions that arise in mixture modeling and multistable systems. In this nonconvex setting, we find that the computational complexity of sampling algorithms scales linearly with the model dimension while that of optimization algorithms scales exponentially.
format Online
Article
Text
id pubmed-6800351
institution National Center for Biotechnology Information
language English
publishDate 2019
publisher National Academy of Sciences
record_format MEDLINE/PubMed
spelling pubmed-68003512019-10-24 Sampling can be faster than optimization Ma, Yi-An Chen, Yuansi Jin, Chi Flammarion, Nicolas Jordan, Michael I. Proc Natl Acad Sci U S A Physical Sciences Optimization algorithms and Monte Carlo sampling algorithms have provided the computational foundations for the rapid growth in applications of statistical machine learning in recent years. There is, however, limited theoretical understanding of the relationships between these 2 kinds of methodology, and limited understanding of relative strengths and weaknesses. Moreover, existing results have been obtained primarily in the setting of convex functions (for optimization) and log-concave functions (for sampling). In this setting, where local properties determine global properties, optimization algorithms are unsurprisingly more efficient computationally than sampling algorithms. We instead examine a class of nonconvex objective functions that arise in mixture modeling and multistable systems. In this nonconvex setting, we find that the computational complexity of sampling algorithms scales linearly with the model dimension while that of optimization algorithms scales exponentially. National Academy of Sciences 2019-10-15 2019-09-30 /pmc/articles/PMC6800351/ /pubmed/31570618 http://dx.doi.org/10.1073/pnas.1820003116 Text en Copyright © 2019 the Author(s). Published by PNAS. https://creativecommons.org/licenses/by-nc-nd/4.0/ https://creativecommons.org/licenses/by-nc-nd/4.0/This open access article is distributed under Creative Commons Attribution-NonCommercial-NoDerivatives License 4.0 (CC BY-NC-ND) (https://creativecommons.org/licenses/by-nc-nd/4.0/) .
spellingShingle Physical Sciences
Ma, Yi-An
Chen, Yuansi
Jin, Chi
Flammarion, Nicolas
Jordan, Michael I.
Sampling can be faster than optimization
title Sampling can be faster than optimization
title_full Sampling can be faster than optimization
title_fullStr Sampling can be faster than optimization
title_full_unstemmed Sampling can be faster than optimization
title_short Sampling can be faster than optimization
title_sort sampling can be faster than optimization
topic Physical Sciences
url https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6800351/
https://www.ncbi.nlm.nih.gov/pubmed/31570618
http://dx.doi.org/10.1073/pnas.1820003116
work_keys_str_mv AT mayian samplingcanbefasterthanoptimization
AT chenyuansi samplingcanbefasterthanoptimization
AT jinchi samplingcanbefasterthanoptimization
AT flammarionnicolas samplingcanbefasterthanoptimization
AT jordanmichaeli samplingcanbefasterthanoptimization