Two-level genetic algorithm for evolving convolutional neural networks for pattern recognition
Artículo
![Thumbnail](/themes/Mirage2/images/cubierta.jpg)
Open/ Download
Access note
Acceso abierto
Publication date
2021Metadata
Show full item record
Cómo citar
Montecino, Daniel A.
Cómo citar
Two-level genetic algorithm for evolving convolutional neural networks for pattern recognition
Abstract
The aim of Neuroevolution is to nd neural networks and convolutional neural network (CNN)
architectures automatically through evolutionary algorithms. A crucial problem in neuroevolution is search
time, since multiple CNNs must be trained during evolution. This problem has led to tness acceleration
approaches, generating a trade-off between time and tness delity. Also, since search spaces for this
problem usually include only a few parameters, this increases the human bias in the search. In this work,
we propose a novel two-level genetic algorithm (GA) for addressing the delity-time trade-off problem
for the tness computation in CNNs. The rst level evaluates many individuals quickly, and the second
evaluates only those with the best results more nely. We also propose a search space with few restrictions,
and an encoding with unexpressed genes to facilitate the crossover operation. This search space allows CNN
architectures to have any sizes, shapes, and skip-connections among nodes. The two-level GA was applied
to the pattern recognition problem on seven datasets, ve MNIST-Variants, Fashion-MNIST, and CIFAR-10,
achieving signi cantly better results than all those previously published. Our results show an improvement
of 39.89% (4.2% error reduction) on the most complex dataset of MNIST (MRDBI), and on average 30.52%
(1.35% error reduction) on all the ve datasets. Furthermore, we show that our algorithm performed as well
as precise-training GA, but took only the time of a fast-training GA. These results can be relevant and useful
not only for image classi cation problems but also for GA-related problems.
Patrocinador
Agencia Nacional de Investigacion y Desarrollo (ANID) FONDECYT 1191610
AFB180004
Department of Electrical Engineering, Universidad de Chile
Advanced Mining Technology Center, Universidad de Chile
Indexation
Artículo de publícación WoS
Quote Item
IEEE Access (2021) Vol. 9 págs. 126856-126872
Collections
The following license files are associated with this item: