A Novel Differential Evolution Based Algorithm for Higher Order Neural Network Training

In this paper, an application of an adaptive differential evolution (DE) algorithm for training higher order neural networks (HONNs), especially the Pi-Sigma Network (PSN) has been introduced. The proposed algorithm is a variant of DE/rand/2/bin and possesses two modifications to avoid the shortcomings of DE/rand/2/bin. The base vector for perturbation is the best vector out of the three randomly selected individuals for mutation, which actually assists intensification keeping the diversification property of DE/rand/2/bin; and novel mutation and crossover strategies are followed considering both exploration and exploitation. The performance of the proposed algorithm for HONN training is evaluated through a wellknown neural network training benchmark i.e. to classify the parity-p problems. The results obtained from the proposed algorithm to train HONN have been compared with solutions from the following algorithms: the basic CRO algorithm, CRO-HONNT and the two most popular variants of the differential evolution algorithm (DE/Rand/1/bin and DE/best/1/bin). It is observed that the application of the proposed algorithm to HONN training (DE-HONNT) performs statistically better than that of other algorithms.