AN MEND PROTOTYPE PATTERN SELECTION ALGORITHM USING GENETIC ALGORITHM

摘要

The selection of prototype plays a decisive part in the performance of synergetic neural network. Amongst the existing prototype pattern selection schemes, the learning algorithm based on information superposition presented by Wang is the most efficient. However, it has a degree parameter greatly affecting the training process to be determined. To overcome this drawback, an improved algorithm is presented and discussed here. This approach makes use of Genetic algorithm, a stochastic search method, to search the global optimum of the unknown parameter in a small search space. Therefore, it converges fairly fast. The experimental results also demonstrate its effectivity.