Randomized algorithms for feedforward neural networks
Fanjun Li, Ying Li · 2016
Randomized algorithm for feedforward neural networks have been summarized and evaluated in this paper. Firstly, a simplified model of feedforward neural networks with random weights is proposed, which consists of a randomized layer and an output layer. Secondly, randomized algorithms for different network structures are summarized on the basis of the simplified model. Finally, several feedforward neural networks with different randomized layers are evaluated on ten UCI data sets. Experiment results show that random vector Functional-link neural network is more powerful than multilayer perception structure with random weights.