HAL will be down for maintenance from Friday, June 10 at 4pm through Monday, June 13 at 9am. More information
Skip to Main content Skip to Navigation
Journal articles

Une nouvelle approche pour les réseaux de neurones: la Représentation Scalaire Distribuée

Abstract : Nowadays, neural networks are largely used in signal and image processing. We propose a new neuron model that uses a special coding of its output, which we will call “Scalar Distributed Representation” (SDR). This representation is based on the idea of representing the neuron’s output by a function, and not only by a scalar. We show that SDR produces a non-linear behaviour of connections between neurons. The SDR is described in general and then specialized on practical considerations. We consider the use of SDRfor a Multi-Layer Perceptron, and we propose a learning algorithm.Finally, we validate the model on two applications : dimensionality reduction, and prediction. In both cases,an important benefit is obtained over the classical model.
Document type :
Journal articles
Complete list of metadata

https://hal.univ-brest.fr/hal-03221467
Contributor : Gilles Burel Connect in order to contact the contributor
Submitted on : Saturday, May 8, 2021 - 7:59:15 PM
Last modification on : Monday, November 22, 2021 - 6:42:03 PM

Identifiers

  • HAL Id : hal-03221467, version 1

Collections

Citation

Gilles Burel. Une nouvelle approche pour les réseaux de neurones: la Représentation Scalaire Distribuée. Traitement du Signal, Lavoisier, 1993, 10 (1), pp.41-51. ⟨hal-03221467⟩

Share

Metrics

Record views

28