An evolutionary genetic neural networks for problems without prior knowledge
Hyoung Uk Ha, Jong‐Kook Kim · 2014
Many problems are now being solved using a version of a neural network (NN). These NN are usually constructed using genetic neural networks (GNNs) for optimizing variables in the NN using a fixed structure or neural evolution (NE) to optimize the structure of the NN using fixed values for the variables in the NN. Thus, previous methods need experienced knowledge of the problem such that either the structure or variables are known to construct a meaningful NN. This paper presents a method called leap evolution adopted neural network (LEANN) that optimizes the NN without prior knowledge such as the values of the variables and the structure of the NN for a given problem. Our method in this paper finds an optimal structure and variables of the NN successfully for the XOR gate problem.