An optimization method of extreme learning machine for regression
Xiaojian Ding, Xiaoguang Liu, Xin Xu · 2016
In this paper, we investigate an optimization scheme for extreme learning machine (ELM) regression, named OELR, to overcome the limitation of ELM that it may lead to overfitting on large training data sets. OELR amounts to minimization of ε-insensitive loss and minimization of the norm of the output weights of single hidden layer feedforward networks (SLFNs). Compared to support vector regression (SVR), OELR has less optimization constraints. Empirical results on the benchmark data sets show that the competitive performance of the OELR over the state-ofthe-art regression learning algorithms.