Numerical approximations for speeding up mcmc inference in the infinite relational model

Mikkel Nørgaard Schmidt; Kristoffer Jon Albers

doi:10.1109/EUSIPCO.2015.7362891

Numerical approximations for speeding up mcmc inference in the infinite relational model

Mikkel Nørgaard Schmidt, Kristoffer Jon Albers

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

505 Downloads (Orbit)

Abstract

The infinite relational model (IRM) is a powerful model for discovering clusters in complex networks; however, the computational speed of Markov chain Monte Carlo inference in the model can be a limiting factor when analyzing large networks. We investigate how using numerical approximations of the log-Gamma function in evaluating the likelihood of the IRM can improve the computational speed of MCMC inference, and how it affects the performance of the model. Using an ensemble of networks generated from the IRM, we compare three approximations in terms of their generalization performance measured on test data. We demonstrate that the computational time for MCMC inference can be reduced by a factor of two without affecting the performance, making it worthwhile in practical situations when on a computational budget.

Original language	English
Title of host publication	Proceedings of the 23rd European Signal Processing Conference (EUSIPCO 2015)
Publisher	IEEE
Publication date	2015
Pages	2781-2785
ISBN (Print)	978-0-9928626-3-3
DOIs	https://doi.org/10.1109/EUSIPCO.2015.7362891
Publication status	Published - 2015
Event	23rd European Signal Processing Conference (EUSIPCO 2015) - Nice, France Duration: 31 Aug 2015 → 4 Sept 2015 Conference number: 23 http://www.eusipco2015.org/

Conference

Conference	23rd European Signal Processing Conference (EUSIPCO 2015)
Number	23
Country/Territory	France
City	Nice
Period	31/08/2015 → 04/09/2015
Internet address	http://www.eusipco2015.org/

Series	Proceedings of the European Signal Processing Conference
ISSN	2076-1465

Keywords

Nonparametric Bayesian modeling
Infinite Relational Model
Numerical approximation

Access to Document

10.1109/EUSIPCO.2015.7362891

Full textAccepted author manuscript, 189 KB

OpenUrl availability

Full text

Cite this

@inproceedings{52afa08457fa4be1bbb3529f50b20ae0,

title = "Numerical approximations for speeding up mcmc inference in the infinite relational model",

abstract = "The infinite relational model (IRM) is a powerful model for discovering clusters in complex networks; however, the computational speed of Markov chain Monte Carlo inference in the model can be a limiting factor when analyzing large networks. We investigate how using numerical approximations of the log-Gamma function in evaluating the likelihood of the IRM can improve the computational speed of MCMC inference, and how it affects the performance of the model. Using an ensemble of networks generated from the IRM, we compare three approximations in terms of their generalization performance measured on test data. We demonstrate that the computational time for MCMC inference can be reduced by a factor of two without affecting the performance, making it worthwhile in practical situations when on a computational budget.",

keywords = "Nonparametric Bayesian modeling, Infinite Relational Model, Numerical approximation",

author = "Schmidt, \{Mikkel N{\o}rgaard\} and Albers, \{Kristoffer Jon\}",

year = "2015",

doi = "10.1109/EUSIPCO.2015.7362891",

language = "English",

isbn = "978-0-9928626-3-3",

series = "Proceedings of the European Signal Processing Conference",

publisher = "IEEE",

pages = "2781--2785",

booktitle = "Proceedings of the 23rd European Signal Processing Conference (EUSIPCO 2015)",

address = "United States",

note = "23rd European Signal Processing Conference (EUSIPCO 2015), EUSIPCO ; Conference date: 31-08-2015 Through 04-09-2015",

url = "http://www.eusipco2015.org/",

}

Schmidt, MN & Albers, KJ 2015, Numerical approximations for speeding up mcmc inference in the infinite relational model. in Proceedings of the 23rd European Signal Processing Conference (EUSIPCO 2015). IEEE, Proceedings of the European Signal Processing Conference, pp. 2781-2785, 23rd European Signal Processing Conference (EUSIPCO 2015), Nice, France, 31/08/2015. https://doi.org/10.1109/EUSIPCO.2015.7362891

Numerical approximations for speeding up mcmc inference in the infinite relational model. / Schmidt, Mikkel Nørgaard; Albers, Kristoffer Jon.
Proceedings of the 23rd European Signal Processing Conference (EUSIPCO 2015). IEEE, 2015. p. 2781-2785 (Proceedings of the European Signal Processing Conference).

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

TY - GEN

T1 - Numerical approximations for speeding up mcmc inference in the infinite relational model

AU - Schmidt, Mikkel Nørgaard

AU - Albers, Kristoffer Jon

N1 - Conference code: 23

PY - 2015

Y1 - 2015

N2 - The infinite relational model (IRM) is a powerful model for discovering clusters in complex networks; however, the computational speed of Markov chain Monte Carlo inference in the model can be a limiting factor when analyzing large networks. We investigate how using numerical approximations of the log-Gamma function in evaluating the likelihood of the IRM can improve the computational speed of MCMC inference, and how it affects the performance of the model. Using an ensemble of networks generated from the IRM, we compare three approximations in terms of their generalization performance measured on test data. We demonstrate that the computational time for MCMC inference can be reduced by a factor of two without affecting the performance, making it worthwhile in practical situations when on a computational budget.

AB - The infinite relational model (IRM) is a powerful model for discovering clusters in complex networks; however, the computational speed of Markov chain Monte Carlo inference in the model can be a limiting factor when analyzing large networks. We investigate how using numerical approximations of the log-Gamma function in evaluating the likelihood of the IRM can improve the computational speed of MCMC inference, and how it affects the performance of the model. Using an ensemble of networks generated from the IRM, we compare three approximations in terms of their generalization performance measured on test data. We demonstrate that the computational time for MCMC inference can be reduced by a factor of two without affecting the performance, making it worthwhile in practical situations when on a computational budget.

KW - Nonparametric Bayesian modeling

KW - Infinite Relational Model

KW - Numerical approximation

U2 - 10.1109/EUSIPCO.2015.7362891

DO - 10.1109/EUSIPCO.2015.7362891

M3 - Article in proceedings

SN - 978-0-9928626-3-3

T3 - Proceedings of the European Signal Processing Conference

SP - 2781

EP - 2785

BT - Proceedings of the 23rd European Signal Processing Conference (EUSIPCO 2015)

PB - IEEE

T2 - 23rd European Signal Processing Conference (EUSIPCO 2015)

Y2 - 31 August 2015 through 4 September 2015

ER -

Numerical approximations for speeding up mcmc inference in the infinite relational model

Abstract

Conference

Keywords

Access to Document

OpenUrl availability

Fingerprint

Cite this