Multi-objective Model Selection for Extreme Learning Machine

Recently, Extreme Learning Machines (ELMs) have get successful application in the fields of classification and regression. However, the generalization performance of ELM will be decreased if there exits non-optimal input weights and hidden biases. To solve this problem, this paper introduced a new model selection method of ELM based on multi-objective optimization. This method views ELM model selection as a multi-objective global optimization problem, in which the generalization error and output weights are as optimization objectives. To accelerate the optimization speed, a fast Leave-one-out (LOO) error estimate of ELM is introduced to refer to the generalization error. Taking into account the contradiction between these two objectives, multi-objective comprehensive learning particle swarm optimization algorithm is utilized to find non-dominated solutions. Experiment on four UCI regression data sets are conducted.