Extreme learning machine-based functional link network

For their universal approximation, more compact topology and faster learning speed, functional-link neural network has attracted considerable attention, and they have been widely applied in many science and engineering fields. Extreme Learning Machine (ELM) algorithm allowing a fast learning speed with high accuracy by randomly generating the input weights of single layer feedforward neural networks, whereas the output weights are analytically determined using the least-square method. In this paper we shows that ELM can be extended to radial basis functional link network (RBFL), which allows the RBFL kernel´s centres and impact widths to be randomly generated after formulating the network in a linear system. Thus, the network weights can be easily solved. The experimental results in artificial and real benchmarking function approximation problems, shows that the RBFL networks with ELM algorithm obtained better results in accuracy, high learning speed and good generalization performance for almost all datasets.