Self-generating neural networks
W.X. Wen, H. Liu, A. Jennings · 2003
A method for generating neural networks automatically is proposed. Not only the weights of the connections but also the structure of the network including the number of neurons, the number of layers, and the interconnections among the neurons are all learned from the training examples. Issues of optimization and pruning of the generated networks are investigated. An experimental system has been implemented based on the proposed method and some experimental results and comparisons between this method and other methods are also given.>