Evolutionary ordered neural network with a linked-list encoding scheme

The paper proposes an evolutionary design of a neural network architecture, with a one dimensional linked list encoding scheme. In this scheme, neurons are arranged in one dimensional array, and the order information of neurons play important roles in genetic operation. Due to one dimensional structure, encoding from neural network architecture to genotype becomes easy, and genetic operation can be easily applied. To avoid the permutation problem, we choose evolutionary programming (EP) rather than genetic algorithm (GA), i.e., we apply mutation operators only in order to generate offspring. The proposed scheme is applied to XOR and 3 parity problems, and optimal neural network architecture can be found with this encoding scheme