Please use this identifier to cite or link to this item: http://hdl.handle.net/10553/107258
Title: Modularity as a means for complexity management in neural networks learning
Authors: Castillo-Bolado, D.
Guerra-Artal, C. 
Hernandez-Tejera, M. 
UNESCO Clasification: 120304 Inteligencia artificial
Issue Date: 2019
Journal: ArXiv.org 
Abstract: Training a Neural Network (NN) with lots of parameters or intricate architectures creates undesired phenomena that complicate the optimization process. To address this issue we propose a first modular approach to NN design, wherein the NN is decomposed into a control module and several functional modules, implementing primitive operations. We illustrate the modular concept by comparing performances between a monolithic and a modular NN on a list sorting problem and show the benefits in terms of training speed, training stability and maintainability. We also discuss some questions that arise in modular NNs.
URI: http://hdl.handle.net/10553/107258
ISSN: 2331-8422
Source: ArXiv.org [ISSN 2331-8422], 25 febrero 2019
Appears in Collections:Artículo preliminar
Thumbnail
Adobe PDF (1,12 MB)
Show full item record

Google ScholarTM

Check


Share



Export metadata



Items in accedaCRIS are protected by copyright, with all rights reserved, unless otherwise indicated.