Probabilistic Forecasting Using Monte Carlo Dropout Neural Networks

Cristián Serpell; Ignacio Araya; Carlos Valle; Héctor Allende

doi:10.1007/978-3-030-33904-3_36

Probabilistic Forecasting Using Monte Carlo Dropout Neural Networks

Cristián Serpell, Ignacio Araya, Carlos Valle, Héctor Allende

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

9 Scopus citations

Abstract

Using artificial neural networks for forecasting tasks is a popular approach that has proven to be very accurate. When used to estimate prediction intervals, a normal distribution is usually assumed as the data noise uncertainty term, as in MVE networks, while model parameters uncertainty is often ignored. Because of this, prediction intervals estimated by them are narrow in uncertain regions where train data is scarce. To tackle this problem we apply Monte Carlo dropout, which is a model uncertainty representation technique, to the network parameters of a Long Short-Term Memory MVE network, allowing us to construct better prediction intervals in probabilistic forecasting tasks. We compare our proposal with the pure MVE method in four wind speed and one consumer load real forecasting datasets, showing that our method improves results in terms of the Winkler loss in both one step ahead and multi-step ahead probabilistic forecasting.

Original language	English
Title of host publication	Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 24th Iberoamerican Congress, CIARP 2019, Proceedings
Editors	Ingela Nyström, Yanio Hernández Heredia, Vladimir Milián Núñez
Publisher	Springer
Pages	387-397
Number of pages	11
ISBN (Print)	9783030339036
DOIs	https://doi.org/10.1007/978-3-030-33904-3_36
State	Published - 2019
Externally published	Yes
Event	24th Iberoamerican Congress on Pattern Recognition, CIARP 2019 - Havana, Cuba Duration: 28 Oct 2019 → 31 Oct 2019

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	11896 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	24th Iberoamerican Congress on Pattern Recognition, CIARP 2019
Country/Territory	Cuba
City	Havana
Period	28/10/19 → 31/10/19

Keywords

Consumer load forecasting
Mean Variance Estimation
Monte Carlo dropout
Prediction interval
Probabilistic forecasting
Wind speed forecasting

Access to Document

10.1007/978-3-030-33904-3_36

Cite this

Serpell, C., Araya, I., Valle, C., & Allende, H. (2019). Probabilistic Forecasting Using Monte Carlo Dropout Neural Networks. In I. Nyström, Y. Hernández Heredia, & V. Milián Núñez (Eds.), Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 24th Iberoamerican Congress, CIARP 2019, Proceedings (pp. 387-397). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11896 LNCS). Springer. https://doi.org/10.1007/978-3-030-33904-3_36

Serpell, Cristián ; Araya, Ignacio ; Valle, Carlos et al. / Probabilistic Forecasting Using Monte Carlo Dropout Neural Networks. Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 24th Iberoamerican Congress, CIARP 2019, Proceedings. editor / Ingela Nyström ; Yanio Hernández Heredia ; Vladimir Milián Núñez. Springer, 2019. pp. 387-397 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{4307da9024ab46dd91e14e0f58ef24ff,

title = "Probabilistic Forecasting Using Monte Carlo Dropout Neural Networks",

abstract = "Using artificial neural networks for forecasting tasks is a popular approach that has proven to be very accurate. When used to estimate prediction intervals, a normal distribution is usually assumed as the data noise uncertainty term, as in MVE networks, while model parameters uncertainty is often ignored. Because of this, prediction intervals estimated by them are narrow in uncertain regions where train data is scarce. To tackle this problem we apply Monte Carlo dropout, which is a model uncertainty representation technique, to the network parameters of a Long Short-Term Memory MVE network, allowing us to construct better prediction intervals in probabilistic forecasting tasks. We compare our proposal with the pure MVE method in four wind speed and one consumer load real forecasting datasets, showing that our method improves results in terms of the Winkler loss in both one step ahead and multi-step ahead probabilistic forecasting.",

keywords = "Consumer load forecasting, Mean Variance Estimation, Monte Carlo dropout, Prediction interval, Probabilistic forecasting, Wind speed forecasting",

author = "Cristi{\'a}n Serpell and Ignacio Araya and Carlos Valle and H{\'e}ctor Allende",

note = "Publisher Copyright: {\textcopyright} Springer Nature Switzerland AG 2019. Copyright: Copyright 2019 Elsevier B.V., All rights reserved.; 24th Iberoamerican Congress on Pattern Recognition, CIARP 2019 ; Conference date: 28-10-2019 Through 31-10-2019",

year = "2019",

doi = "10.1007/978-3-030-33904-3_36",

language = "English",

isbn = "9783030339036",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer",

pages = "387--397",

editor = "Ingela Nystr{\"o}m and {Hern{\'a}ndez Heredia}, Yanio and {Mili{\'a}n N{\'u}{\~n}ez}, Vladimir",

booktitle = "Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 24th Iberoamerican Congress, CIARP 2019, Proceedings",

}

Serpell, C, Araya, I, Valle, C & Allende, H 2019, Probabilistic Forecasting Using Monte Carlo Dropout Neural Networks. in I Nyström, Y Hernández Heredia & V Milián Núñez (eds), Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 24th Iberoamerican Congress, CIARP 2019, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11896 LNCS, Springer, pp. 387-397, 24th Iberoamerican Congress on Pattern Recognition, CIARP 2019, Havana, Cuba, 28/10/19. https://doi.org/10.1007/978-3-030-33904-3_36

Probabilistic Forecasting Using Monte Carlo Dropout Neural Networks. / Serpell, Cristián; Araya, Ignacio; Valle, Carlos et al.
Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 24th Iberoamerican Congress, CIARP 2019, Proceedings. ed. / Ingela Nyström; Yanio Hernández Heredia; Vladimir Milián Núñez. Springer, 2019. p. 387-397 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11896 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Probabilistic Forecasting Using Monte Carlo Dropout Neural Networks

AU - Serpell, Cristián

AU - Araya, Ignacio

AU - Valle, Carlos

AU - Allende, Héctor

PY - 2019

Y1 - 2019

N2 - Using artificial neural networks for forecasting tasks is a popular approach that has proven to be very accurate. When used to estimate prediction intervals, a normal distribution is usually assumed as the data noise uncertainty term, as in MVE networks, while model parameters uncertainty is often ignored. Because of this, prediction intervals estimated by them are narrow in uncertain regions where train data is scarce. To tackle this problem we apply Monte Carlo dropout, which is a model uncertainty representation technique, to the network parameters of a Long Short-Term Memory MVE network, allowing us to construct better prediction intervals in probabilistic forecasting tasks. We compare our proposal with the pure MVE method in four wind speed and one consumer load real forecasting datasets, showing that our method improves results in terms of the Winkler loss in both one step ahead and multi-step ahead probabilistic forecasting.

AB - Using artificial neural networks for forecasting tasks is a popular approach that has proven to be very accurate. When used to estimate prediction intervals, a normal distribution is usually assumed as the data noise uncertainty term, as in MVE networks, while model parameters uncertainty is often ignored. Because of this, prediction intervals estimated by them are narrow in uncertain regions where train data is scarce. To tackle this problem we apply Monte Carlo dropout, which is a model uncertainty representation technique, to the network parameters of a Long Short-Term Memory MVE network, allowing us to construct better prediction intervals in probabilistic forecasting tasks. We compare our proposal with the pure MVE method in four wind speed and one consumer load real forecasting datasets, showing that our method improves results in terms of the Winkler loss in both one step ahead and multi-step ahead probabilistic forecasting.

KW - Consumer load forecasting

KW - Mean Variance Estimation

KW - Monte Carlo dropout

KW - Prediction interval

KW - Probabilistic forecasting

KW - Wind speed forecasting

UR - http://www.scopus.com/inward/record.url?scp=85075682649&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-33904-3_36

DO - 10.1007/978-3-030-33904-3_36

M3 - Conference contribution

AN - SCOPUS:85075682649

SN - 9783030339036

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 387

EP - 397

BT - Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 24th Iberoamerican Congress, CIARP 2019, Proceedings

A2 - Nyström, Ingela

A2 - Hernández Heredia, Yanio

A2 - Milián Núñez, Vladimir

PB - Springer

T2 - 24th Iberoamerican Congress on Pattern Recognition, CIARP 2019

Y2 - 28 October 2019 through 31 October 2019

ER -

Serpell C, Araya I, Valle C, Allende H. Probabilistic Forecasting Using Monte Carlo Dropout Neural Networks. In Nyström I, Hernández Heredia Y, Milián Núñez V, editors, Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 24th Iberoamerican Congress, CIARP 2019, Proceedings. Springer. 2019. p. 387-397. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-33904-3_36

Probabilistic Forecasting Using Monte Carlo Dropout Neural Networks

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this