Meta-Learning Evolutionary Artificial Neural Networks: by Means of Cellular Automata

In this paper, we introduce meta-learning evolutionary artificial neural network by means of cellular automata (MLEANN-CA). It is an adaptive computational framework based on evolutionary computation with indirect encoding methods (cellular automata) for automatic design of optimal artificial neural networks wherein the neural network architecture, activation function, connection weights, and the learning algorithm with its parameters are adapted according to the problem. We explored, experimentally, the performance of the MLEANN-CA framework using the Matlab and two famous chaotic time series. We compared it with a previous MLEANN framework, which used the direct encoding methods, and with the conventional design of ANNs. We demonstrated how effective is the proposed MLEANN-CA framework to obtain a design of feed-forward neural network that is smaller, faster and with better generalization performance