Efficient sparse matrix-delayed vector multiplication for
discretized neural field model

J 2018

Efficient sparse matrix-delayed vector multiplication for discretized neural field model

FOUSEK, Jan

Basic information

Original name

Efficient sparse matrix-delayed vector multiplication for discretized neural field model

Authors

FOUSEK, Jan (203 Czech Republic, guarantor, belonging to the institution)

Edition

The Journal of Supercomputing, Springer US, 2018, 0920-8542

Other information

Language

English

Type of outcome

Article in a journal

Field of Study

10201 Computer sciences, information science, bioinformatics

Country of publisher

United States of America

Confidentiality degree

is not subject to a state or trade secret

References:

URL

Impact factor

Impact factor: 2.157

RIV identification code

RIV/00216224:14610/18:00102136

Organization unit

Institute of Computer Science

DOI

http://dx.doi.org/10.1007/s11227-017-2194-4

UT WoS

000430412400005

EID Scopus

2-s2.0-85038114439

Keywords in English

Neural field;Sparse matrix;SpMV;Delay differential equations;Data locality

Abstract

In the original language

Computational models of the human brain provide an important tool for studying the principles behind brain function and disease. To achieve whole-brain simulation, models are formulated at the level of neuronal populations as systems of delayed differential equations. In this paper, we show that the integration of large systems of sparsely connected neural masses is similar to well-studied sparse matrix-vector multiplication; however, due to delayed contributions, it differs in the data access pattern to the vectors. To improve data locality, we propose a combination of node reordering and tiled schedules derived from the connectivity matrix of the particular system, which allows performing multiple integration steps within a tile. We present two schedules: with a serial processing of the tiles and one allowing for parallel processing of the tiles. We evaluate the presented schedules showing speedup up to 2x on single-socket CPU, and 1.25x on Xeon Phi accelerator.

Links

EF16_013/0001802, research and development project

Name: CERIT Scientific Cloud

MUNI/A/0897/2016, interní kód MU

Name: Rozsáhlé výpočetní systémy: modely, aplikace a verifikace VI.

Investor: Masaryk University, Category A

Cite

FOUSEK, Jan. Efficient sparse matrix-delayed vector multiplication for discretized neural field model. The Journal of Supercomputing. Springer US, 2018, vol. 74, No 5, p. 1863-1884. ISSN 0920-8542. Available from: https://dx.doi.org/10.1007/s11227-017-2194-4.

@article{1399331,
   author = {Fousek, Jan},
   article_number = {5},
   doi = {http://dx.doi.org/10.1007/s11227-017-2194-4},
   keywords = {Neural field;Sparse matrix;SpMV;Delay differential equations;Data locality},
   language = {eng},
   issn = {0920-8542},
   journal = {The Journal of Supercomputing},
   title = {Efficient sparse matrix-delayed vector multiplication for discretized neural field model},
   url = {https://doi.org/10.1007/s11227-017-2194-4},
   volume = {74},
   year = {2018}
}

TY  - JOUR
ID  - 1399331
AU  - Fousek, Jan
PY  - 2018
TI  - Efficient sparse matrix-delayed vector multiplication for discretized neural field model
JF  - The Journal of Supercomputing
VL  - 74
IS  - 5
SP  - 1863-1884
EP  - 1863-1884
PB  - Springer US
SN  - 09208542
KW  - Neural field;Sparse matrix;SpMV;Delay differential equations;Data locality
UR  - https://doi.org/10.1007/s11227-017-2194-4
L2  - https://doi.org/10.1007/s11227-017-2194-4
N2  - Computational models of the human brain provide an important tool for studying the principles behind brain function and disease. To achieve whole-brain simulation, models are formulated at the level of neuronal populations as systems of delayed differential equations. In this paper, we show that the integration of large systems of sparsely connected neural masses is similar to well-studied sparse matrix-vector multiplication; however, due to delayed contributions, it differs in the data access pattern to the vectors. To improve data locality, we propose a combination of node reordering and tiled schedules derived from the connectivity matrix of the particular system, which allows performing multiple integration steps within a tile. We present two schedules: with a serial processing of the tiles and one allowing for parallel processing of the tiles. We evaluate the presented schedules showing speedup up to 2x on single-socket CPU, and 1.25x on Xeon Phi accelerator.
ER  -

FOUSEK, Jan. Efficient sparse matrix-delayed vector multiplication for discretized neural field model. \textit{The Journal of Supercomputing}. Springer US, 2018, vol.~74, No~5, p.~1863-1884. ISSN~0920-8542. Available from: https://dx.doi.org/10.1007/s11227-017-2194-4.

Přehled o publikaci