Subnet Weight Modification Algorithm for Ensemble

摘要

In view of comparability between neural network ensemble and Adaline, a modification algorithm for ensemble weights is presented based on the gradient descent method. This algorithm can improve the generalization performance by modifying subnet weights after the ensemble subnets are trained individually. Simulation results indicate that the new algorithm is of subnet selection function similar to GASEN but on a different idea, and of better performance than single network, simple-averaged ensemble and GASEN.