Double-paralleled ridgelet neural network with IFPSO training algorithm

To speed up the convergence and promote the generalized performance of adaptive ridgelet neural network, we present a new model, Double-paralleled Ridgelet Neural Network, which consists of two paralleled networks -a hidden-layer adaptive ridgelet network and a single-layer feedforward neural network. In order to obtain higher accuracy and learning speed, regardless of the curses of nonlinear parameters in ridgelet activation function, an improved flock-of-starling particle swarm optimization algorithm is introduced as the training algorithm, which is able to converge on the global minimum by means of two dissimilar measurements with FPSO - adaptive inertia weights and near-neighbored topological interactions. The classification experiments indicate that the new model has better classification performance and simple structure compared with conventional classifiers RBF and SVM.