Modularity as a Means for Complexity Management in Neural Networks Learning
Machine Learning
2019-02-26 v1 Neural and Evolutionary Computing
Machine Learning
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.
Cite
@article{arxiv.1902.09240,
title = {Modularity as a Means for Complexity Management in Neural Networks Learning},
author = {David Castillo-Bolado and Cayetano Guerra-Artal and Mario Hernandez-Tejera},
journal= {arXiv preprint arXiv:1902.09240},
year = {2019}
}
Comments
Full-paper submited to the AAAI-MAKE 2019