The self organizing neural network algorithm: adapting structure for optimum supervised learning

An algorithm called the self-organizing neural network (SONN) is described, and its use as a supervised learning architecture is demonstrated. The algorithm constructs a network, chooses the neuron functions, and adjusts the weights. The final network structure is optimal in the sense that it uses simulated annealing in the model search. The results (number of weights, complexity of the final structure, computer time, and model accuracy) are compared to the back-propagation algorithm. They show that SONN constructs a simpler, more accurate model, requiring fewer training data and epochs