An interval particle swarm optimization algorithm for evolving interval neural networks
Shouping Guan, Jingjing Zhang, Zou Li-fu · 2018
Considering the problems that the neural networks always fall into local optimum when they are trained by the gradient descent method, this paper proposes an interval particle swarm optimization algorithm (IPSO) to evolve the parameters of the interval neural networks (INNs), which can express the uncertain information of systems. The IPSO is a real-valued particle swarm optimization algorithm combined with the interval analysis, which can evolve the interval parameters for the interval models. The simulation results show that the IPSO algorithm can evolve the INNs with satisfied accuracy for the testing interval numerical functions, and provides a new way for the training of INNs.