Abstract
In this report is presented a first experiment of a parallel application implemented on a Quadrics supercomputer in ENEA (Italian National Agency for New Technologies, Energy and the Environment). On this SIMD parallel machine it has been realized a learning algorithm for multilayer perceptrons, the so called backpropagation. This feasibility study has been aimed towards the solving of the problem of the logical exclusive or (XOR). The main reason being the nonlinear separability of such a problem, i.e. the need for a hidden layer in the neural architecture. The parallel algorithm is presented together with its performances under different optimizing techniques on Quadrics.
Taraglio, S
[1]
- ENEA, Casaccia (Italy). Area Energia e Innovazione
Citation Formats
Taraglio, S.
First experiments of parallel application in ENEA (Italian National Agency for New Technologies, Energy and the Environment): Neural networks; Note sui primi esperimenti di supercalcolo parallelo in ENEA: Le reti neurali.
Italy: N. p.,
1994.
Web.
Taraglio, S.
First experiments of parallel application in ENEA (Italian National Agency for New Technologies, Energy and the Environment): Neural networks; Note sui primi esperimenti di supercalcolo parallelo in ENEA: Le reti neurali.
Italy.
Taraglio, S.
1994.
"First experiments of parallel application in ENEA (Italian National Agency for New Technologies, Energy and the Environment): Neural networks; Note sui primi esperimenti di supercalcolo parallelo in ENEA: Le reti neurali."
Italy.
@misc{etde_10123641,
title = {First experiments of parallel application in ENEA (Italian National Agency for New Technologies, Energy and the Environment): Neural networks; Note sui primi esperimenti di supercalcolo parallelo in ENEA: Le reti neurali}
author = {Taraglio, S}
abstractNote = {In this report is presented a first experiment of a parallel application implemented on a Quadrics supercomputer in ENEA (Italian National Agency for New Technologies, Energy and the Environment). On this SIMD parallel machine it has been realized a learning algorithm for multilayer perceptrons, the so called backpropagation. This feasibility study has been aimed towards the solving of the problem of the logical exclusive or (XOR). The main reason being the nonlinear separability of such a problem, i.e. the need for a hidden layer in the neural architecture. The parallel algorithm is presented together with its performances under different optimizing techniques on Quadrics.}
place = {Italy}
year = {1994}
month = {Nov}
}
title = {First experiments of parallel application in ENEA (Italian National Agency for New Technologies, Energy and the Environment): Neural networks; Note sui primi esperimenti di supercalcolo parallelo in ENEA: Le reti neurali}
author = {Taraglio, S}
abstractNote = {In this report is presented a first experiment of a parallel application implemented on a Quadrics supercomputer in ENEA (Italian National Agency for New Technologies, Energy and the Environment). On this SIMD parallel machine it has been realized a learning algorithm for multilayer perceptrons, the so called backpropagation. This feasibility study has been aimed towards the solving of the problem of the logical exclusive or (XOR). The main reason being the nonlinear separability of such a problem, i.e. the need for a hidden layer in the neural architecture. The parallel algorithm is presented together with its performances under different optimizing techniques on Quadrics.}
place = {Italy}
year = {1994}
month = {Nov}
}