A Novel Back-propagation Neural Network Training Algorithm Designed by an Ant Colony Optimization

This article presents a new back-propagation neural network (BPN) training algorithm performed with an ant colony optimization (ACO) to get the optimal connection weights of the BPN. The concentration of pheromone laid by the artificial ants moving on the connection path is the key factor of the optimal weight determination. This is the metaphor that the optimal weights make the "total length of neuron-to-neuron connections traversed by all artificial ants," which is defined by the BPN output values and the target values of the proposed ant-based BPN, be the shortest. The generalization ability of the proposed ant-based BPN was checked by means of approximating two functions, a statistic standard normal distribution N(0,1) and a Sin(x) function, and demonstrated by comparing it to that of a standard BPN. Surprisingly, the superiority of the ant-based BPN\´s generalization ability in terms of both the statistical determination coefficient and the means square error is beyond the expectation