Optimization and Performance Analysis of Extreme Learning Machine by L2-Norm Regularization

Extreme learning machine is a new feedforward neural network. Compared with the traditional neural network, it has the advantages of simple network construction and fast learning. However, as the least square method is used to solve the optimal output weight, the extreme learning machine has some problems such as weak anti-interference ability, poor stability and over-fitting. For the above problems, L2-norm regularization is adopted to optimize the extreme learning machine model, and minimize the training error and output weight by determining the regular parameters in this paper. Simulation results on standard data sets shows that optimization model by L2-norm regularization can significantly improve the stability and anti-interference of model.