Bayesian Leave-One-Out Cross-Validation for Large Data

Michael Riis Andersen; Mans  Magnusson; Johan Jonasson; Aki Vehtari

Bayesian Leave-One-Out Cross-Validation for Large Data

Michael Riis Andersen
, Mans Magnusson
, Johan Jonasson
, Aki Vehtari

Aalto University
Chalmers University of Technology

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

499 Downloads (Orbit)

Abstract

Model inference, such as model comparison, model checking, and model selection, is an important part of model development. Leave-one-out cross-validation (LOO-CV) is a general approach for assessing the generalizability of a model, but unfortunately, LOO-CV does not scale well to large datasets. We propose a combination of using approximate inference techniques and probabilityproportional-to-size-sampling (PPS) for fast LOOCV model evaluation for large data. We provide both theoretical and empirical results showing good properties for large data.

Original language	English
Title of host publication	Proceedings of the 36^th International Conference on Machine Learning
Publisher	International Machine Learning Society (IMLS)
Publication date	2019
Pages	7505-7525
ISBN (Print)	9781510886988
Publication status	Published - 2019
Event	36th International Conference on Machine Learning - Long Beach Convention Center, Long Beach, United States Duration: 10 Jun 2019 → 15 Jun 2019 Conference number: 36

Conference

Conference	36th International Conference on Machine Learning
Number	36
Location	Long Beach Convention Center
Country/Territory	United States
City	Long Beach
Period	10/06/2019 → 15/06/2019

Access to Document

FulltextFinal published version, 283 KB

OpenUrl availability

Check availability in DTU Findit

Cite this

@inproceedings{450192a8805849aea97edf854fd23748,

title = "Bayesian Leave-One-Out Cross-Validation for Large Data",

abstract = "Model inference, such as model comparison, model checking, and model selection, is an important part of model development. Leave-one-out cross-validation (LOO-CV) is a general approach for assessing the generalizability of a model, but unfortunately, LOO-CV does not scale well to large datasets. We propose a combination of using approximate inference techniques and probabilityproportional-to-size-sampling (PPS) for fast LOOCV model evaluation for large data. We provide both theoretical and empirical results showing good properties for large data. ",

author = "Andersen, \{Michael Riis\} and Mans Magnusson and Johan Jonasson and Aki Vehtari",

year = "2019",

language = "English",

isbn = "9781510886988",

pages = "7505--7525",

booktitle = "Proceedings of the 36th International Conference on Machine Learning",

publisher = "International Machine Learning Society (IMLS)",

note = "36th International Conference on Machine Learning, ICML 2019 ; Conference date: 10-06-2019 Through 15-06-2019",

}

Bayesian Leave-One-Out Cross-Validation for Large Data. / Andersen, Michael Riis; Magnusson, Mans ; Jonasson, Johan et al.
Proceedings of the 36^th International Conference on Machine Learning. International Machine Learning Society (IMLS), 2019. p. 7505-7525.

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

TY - GEN

T1 - Bayesian Leave-One-Out Cross-Validation for Large Data

AU - Andersen, Michael Riis

AU - Magnusson, Mans

AU - Jonasson, Johan

AU - Vehtari, Aki

N1 - Conference code: 36

PY - 2019

Y1 - 2019

N2 - Model inference, such as model comparison, model checking, and model selection, is an important part of model development. Leave-one-out cross-validation (LOO-CV) is a general approach for assessing the generalizability of a model, but unfortunately, LOO-CV does not scale well to large datasets. We propose a combination of using approximate inference techniques and probabilityproportional-to-size-sampling (PPS) for fast LOOCV model evaluation for large data. We provide both theoretical and empirical results showing good properties for large data.

AB - Model inference, such as model comparison, model checking, and model selection, is an important part of model development. Leave-one-out cross-validation (LOO-CV) is a general approach for assessing the generalizability of a model, but unfortunately, LOO-CV does not scale well to large datasets. We propose a combination of using approximate inference techniques and probabilityproportional-to-size-sampling (PPS) for fast LOOCV model evaluation for large data. We provide both theoretical and empirical results showing good properties for large data.

M3 - Article in proceedings

SN - 9781510886988

SP - 7505

EP - 7525

BT - Proceedings of the 36th International Conference on Machine Learning

PB - International Machine Learning Society (IMLS)

T2 - 36th International Conference on Machine Learning

Y2 - 10 June 2019 through 15 June 2019

ER -

Bayesian Leave-One-Out Cross-Validation for Large Data

Abstract

Conference

Access to Document

OpenUrl availability

Fingerprint

Cite this