Evolutionary projection neural networks
Min Woong Hwang, Jin Young Choi, Jae‐Hong Park · 2002
Presents an evolutionary projection neural network (PNN) trained by an evolutionary computation technique. The PNN can activate radial basis functions as well as sigmoid functions with a special type of hidden nodes. The evolutionary learning algorithm for the PNN not only trains the parameters and the connection weights but also optimizes the network structure. The structure optimization strategy not only determines the number of hidden nodes necessary to represent a given target function, but also decides whether the role of each hidden node is a radial basis function node or a sigmoid function node. In order to apply the algorithm, a PNN is realized by a self-organizing genotype representation with a linked-list data structure. Simulations show that the algorithm can build a PNN with less hidden nodes than are required by the existing learning algorithm, which uses error backpropagation (EBP) and the network growing strategy.