MPI/OpenMP Hybrid Parallel Inference Methods for Latent Dirichlet Allocation – Approximation and Evaluation

Shotaro TORA; Koji EGUCHI

doi:10.1587/transinf.E96.D.1006

MPI/OpenMP Hybrid Parallel Inference Methods for Latent Dirichlet Allocation – Approximation and Evaluation

Shotaro TORA, Koji EGUCHI

Full Text Views

0

Cite this

Summary :

Recently, probabilistic topic models have been applied to various types of data, including text, and their effectiveness has been demonstrated. Latent Dirichlet allocation (LDA) is a well known topic model. Variational Bayesian inference or collapsed Gibbs sampling is often used to estimate parameters in LDA; however, these inference methods incur high computational cost for large-scale data. Therefore, highly efficient technology is needed for this purpose. We use parallel computation technology for efficient collapsed Gibbs sampling inference for LDA. We assume a symmetric multiprocessing (SMP) cluster, which has been widely used in recent years. In prior work on parallel inference for LDA, either MPI or OpenMP has often been used alone. For an SMP cluster, however, it is more suitable to adopt hybrid parallelization that uses message passing for communication between SMP nodes and loop directives for parallelization within each SMP node. We developed an MPI/OpenMP hybrid parallel inference method for LDA, and evaluated the performance of the inference under various settings of an SMP cluster. We further investigated the approximation that controls the inter-node communications, and found out that it achieved noticeable increase in inference speed while maintaining inference accuracy.

Publication: IEICE TRANSACTIONS on Information Vol.E96-D No.5 pp.1006-1015

Publication Date: 2013/05/01

Publicized

Online ISSN: 1745-1361

DOI: 10.1587/transinf.E96.D.1006

Type of Manuscript: Special Section PAPER (Special Section on Data Engineering and Information Management)

Category: Advanced Search

Cite this

Copy

Shotaro TORA, Koji EGUCHI, "MPI/OpenMP Hybrid Parallel Inference Methods for Latent Dirichlet Allocation – Approximation and Evaluation" in IEICE TRANSACTIONS on Information, vol. E96-D, no. 5, pp. 1006-1015, May 2013, doi: 10.1587/transinf.E96.D.1006.
Abstract: Recently, probabilistic topic models have been applied to various types of data, including text, and their effectiveness has been demonstrated. Latent Dirichlet allocation (LDA) is a well known topic model. Variational Bayesian inference or collapsed Gibbs sampling is often used to estimate parameters in LDA; however, these inference methods incur high computational cost for large-scale data. Therefore, highly efficient technology is needed for this purpose. We use parallel computation technology for efficient collapsed Gibbs sampling inference for LDA. We assume a symmetric multiprocessing (SMP) cluster, which has been widely used in recent years. In prior work on parallel inference for LDA, either MPI or OpenMP has often been used alone. For an SMP cluster, however, it is more suitable to adopt hybrid parallelization that uses message passing for communication between SMP nodes and loop directives for parallelization within each SMP node. We developed an MPI/OpenMP hybrid parallel inference method for LDA, and evaluated the performance of the inference under various settings of an SMP cluster. We further investigated the approximation that controls the inter-node communications, and found out that it achieved noticeable increase in inference speed while maintaining inference accuracy.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.E96.D.1006/_p

Copy

@ARTICLE{e96-d_5_1006,
author={Shotaro TORA, Koji EGUCHI, },
journal={IEICE TRANSACTIONS on Information},
title={MPI/OpenMP Hybrid Parallel Inference Methods for Latent Dirichlet Allocation – Approximation and Evaluation},
year={2013},
volume={E96-D},
number={5},
pages={1006-1015},
abstract={Recently, probabilistic topic models have been applied to various types of data, including text, and their effectiveness has been demonstrated. Latent Dirichlet allocation (LDA) is a well known topic model. Variational Bayesian inference or collapsed Gibbs sampling is often used to estimate parameters in LDA; however, these inference methods incur high computational cost for large-scale data. Therefore, highly efficient technology is needed for this purpose. We use parallel computation technology for efficient collapsed Gibbs sampling inference for LDA. We assume a symmetric multiprocessing (SMP) cluster, which has been widely used in recent years. In prior work on parallel inference for LDA, either MPI or OpenMP has often been used alone. For an SMP cluster, however, it is more suitable to adopt hybrid parallelization that uses message passing for communication between SMP nodes and loop directives for parallelization within each SMP node. We developed an MPI/OpenMP hybrid parallel inference method for LDA, and evaluated the performance of the inference under various settings of an SMP cluster. We further investigated the approximation that controls the inter-node communications, and found out that it achieved noticeable increase in inference speed while maintaining inference accuracy.},
keywords={},
doi={10.1587/transinf.E96.D.1006},
ISSN={1745-1361},
month={May},}

Copy

TY - JOUR
TI - MPI/OpenMP Hybrid Parallel Inference Methods for Latent Dirichlet Allocation – Approximation and Evaluation
T2 - IEICE TRANSACTIONS on Information
SP - 1006
EP - 1015
AU - Shotaro TORA
AU - Koji EGUCHI
PY - 2013
DO - 10.1587/transinf.E96.D.1006
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E96-D
IS - 5
JA - IEICE TRANSACTIONS on Information
Y1 - May 2013
AB - Recently, probabilistic topic models have been applied to various types of data, including text, and their effectiveness has been demonstrated. Latent Dirichlet allocation (LDA) is a well known topic model. Variational Bayesian inference or collapsed Gibbs sampling is often used to estimate parameters in LDA; however, these inference methods incur high computational cost for large-scale data. Therefore, highly efficient technology is needed for this purpose. We use parallel computation technology for efficient collapsed Gibbs sampling inference for LDA. We assume a symmetric multiprocessing (SMP) cluster, which has been widely used in recent years. In prior work on parallel inference for LDA, either MPI or OpenMP has often been used alone. For an SMP cluster, however, it is more suitable to adopt hybrid parallelization that uses message passing for communication between SMP nodes and loop directives for parallelization within each SMP node. We developed an MPI/OpenMP hybrid parallel inference method for LDA, and evaluated the performance of the inference under various settings of an SMP cluster. We further investigated the approximation that controls the inter-node communications, and found out that it achieved noticeable increase in inference speed while maintaining inference accuracy.
ER -