Improved PSO Algorithm for Training of Neural Network in Co-design Architecture

This paper proposes a new version of the standard particle swarm optimization (SPSO) algorithm to train a neural network (NN). The improved PSO, called the wPSOd−CV algorithm, is the improved version of the PSOd−CV algorithm presented in a previous study. The wPSOd−CV algorithm is introduced to solve the issue of premature convergence of the SPSO algorithm. The proposed wPSOd−CV algorithm is used in a co-design architecture. Experimental results confirmed the effectiveness of the NN trained by the wPSOd−CV algorithm when compared with the NN trained by the SPSO algorithm and the PSOd−CV algorithm concerning the minimum learning error and the recognition rates.