|Title:||Modularity as a means for complexity management in neural networks learning||Authors:||Castillo-Bolado, D.
|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|
Items in accedaCRIS are protected by copyright, with all rights reserved, unless otherwise indicated.