Hybrid extreme learning machine approach for homogeneous neural networks

In this study, we propose a novel hybrid structure method called a structured composite model for creating a series of custom neurons using different neuron subunits. The hybrid structure is supervised by a control structure called a homogeneous hybrid extreme learning machine (Ho-HyELM), which creates a series of homogeneous single-layer neural networks using these custom neurons, where each has a different number of hidden units. These networks are trained with the extreme learning machine (ELM) algorithm. The proposed Ho-HyELM approach was applied to a series of regression and classification problems, and the results obtained indicate that the proposed method for splitting a neuron into neuron subunits creates optimal different network types for each problem. The custom ELM-trained networks are more optimal than the commonly used linear unit networks with the sigmoid transfer function.