Busca por arquiteturas de redes neurais artificiais profundas utilizando algoritmos genéticos
Artificial Neural Networks are the pillars of advances in the last two decades in the field of Artificial Intelligence. However, success in training, aiming to obtain good results, strongly depends on the choice of values of a set of hyperparameters associated with their construction. The choice of...
| Autor principal: | RIBEIRO, Denys Menfredy Ferreira |
|---|---|
| Grau: | Artigo |
| Publicado em: |
2023
|
| Assuntos: | |
| Acesso em linha: |
https://bdm.ufpa.br:8443/jspui/handle/prefix/5259 |
| Resumo: |
|---|
|
Artificial Neural Networks are the pillars of advances in the last two decades in the field of Artificial Intelligence. However, success in training, aiming to obtain good results, strongly depends on the choice of values of a set of hyperparameters associated with their construction. The choice of these parameters is usually done empirically, wasting time and generating costs. In this work, it is proposed the development of an ENAS (Evolutionary Neural Architecture Search) algorithm using genetic algorithms as a search method to automate the process of designing architectures of convolutional neural network architectures applied to the task of classifying images from the CIFAR-10 dataset. The proposed algorithm was able to find architectures that obtained good accuracy results in the test set. |