Evolving artificial neural network using simple augmenting weight matrix method

In this paper, a simple method is proposed to evolve artificial neural networks(ANNs) using augmenting weight matrix method(AWMM). ANNs´ architecture and connection weights can be evolved simultaneously by AWMM, and their structures incrementally are growing up from minimal structure. It is a non-mating method. It employs 5 mutation operators: add connection, add node, delete connection, delete node, and new initial weight. And the connection weight is trained by the simplified alopex method, which is a correlation based method for solving optimization problem. In AWMM, structural information is encoded to weighting matrix, and the matrix is augmenting as the hidden nodes are added.