A novel multi-objective grammar-based framework for the generation of Convolutional Neural Networks

Cleber A. C. F. da Silva, Daniel Carneiro Rosa, Péricles B. C. de Miranda, Filipe R. Cordeiro, Tapas Si, André C. A. Nascimento, Rafael Ferreira Leite de Mello, Paulo S. G. de Mattos Neto

Abstract: Highlights•An evolutional framework that optimizes CNNs with no need of an expert is introduced.•A context-free grammar responsible for modeling the architecture of CNNs is proposed.•The framework is tested on 5 datasets, guided by two metrics (accuracy and F1-score).•The results obtained are analyzed and compared to other known architectures.•The framework is capable of generating models that surpass other known architectures.