Resumo

ROJAS DELGADO, Jairo  e  TRUJILLO RASUA, Rafael. Firefly meta-heuristic algorithm applied to artificial neural network pre-training. Rev cuba cienc informat [online]. 2018, vol.12, n.1, pp.14-27. ISSN 2227-1899.

Artificial Neural Networks (ANN) is an active research topic in the scientific community. Firefly algorithm has been successfully used on pre-training ANNs, aiming to avoid local minima convergence of conventional training methods such as Stochastic Gradient Descent. However, pre-training ANNs with a high number of parameters using Firefly algorithms, or any other metaheuristic optimization technique, usually is a high computationally complex task. This paper presents a variation of the firefly algorithm that trains the ANN with a subset of training patterns without causing negative effects in the convergence.

Palavras-chave : artificial neural network; deep learning; firefly; local minima; pre-training.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )