High-performance Implementation of Recurrent Neural Networks on Graphics Processing Units

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F10%3A00177603" target="_blank" >RIV/68407700:21230/10:00177603 - isvavai.cz</a>

Výsledek na webu

—

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

High-performance Implementation of Recurrent Neural Networks on Graphics Processing Units

Popis výsledku v původním jazyce

The evolutionary algorithms work with large sets of individuals. In our case these individuals represent neural networks. The speed of evaluation of the networks is the very crucial factor because it affects the overall speed of the whole evolution process. In this paper we describe our implementation of the fully recurrent neural networks on the general-purpose graphics processing units. We are using the nVidia CUDA technology to accelerate the simulation of the population of the networks. We have created package for Wolfram Mathematica that provides interface to our accelerated simulator from high-level programming environment. Our library supports the client-server architecture, so you can run the simulations on dedicated CUDA-enabled computationalserver and process the results of the simulations on your desktop using the TCP/IP communication protocol. In this paper we present the results of the speedup experiments.

Název v anglickém jazyce

High-performance Implementation of Recurrent Neural Networks on Graphics Processing Units

Popis výsledku anglicky

The evolutionary algorithms work with large sets of individuals. In our case these individuals represent neural networks. The speed of evaluation of the networks is the very crucial factor because it affects the overall speed of the whole evolution process. In this paper we describe our implementation of the fully recurrent neural networks on the general-purpose graphics processing units. We are using the nVidia CUDA technology to accelerate the simulation of the population of the networks. We have created package for Wolfram Mathematica that provides interface to our accelerated simulator from high-level programming environment. Our library supports the client-server architecture, so you can run the simulations on dedicated CUDA-enabled computationalserver and process the results of the simulations on your desktop using the TCP/IP communication protocol. In this paper we present the results of the speedup experiments.

Druh

D - Stať ve sborníku

CEP obor

IN - Informatika

OECD FORD obor

—

Projekt

—

Návaznosti

Z - Vyzkumny zamer (s odkazem do CEZ)

Rok uplatnění

2010

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Proceedings of the 7th EUROSIM Congress on Modelling and Simulation, Vol. 2: Full Papers

ISBN

978-80-01-04589-3

ISSN

—

e-ISSN

—

Počet stran výsledku

4

Strana od-do

—

Název nakladatele

Department of Computer Science and Engineering, FEE, CTU in Prague

Místo vydání

Prague

Místo konání akce

Praha

Datum konání akce

6. 9. 2010

Typ akce podle státní příslušnosti

EUR - Evropská akce

Kód UT WoS článku

—