A modified ELM algorithm for single-hidden layer feedforward neural networks with linear nodes

A modified ELM algorithm for a class of single-hidden layer feedforward neural networks (SLFNs) with linear nodes is discussed in this paper. It is seen that the input weights of the SLFN are designed such that the hidden layer performs as a preprocessor for removing the effects of the input disturbance and reducing both the structural and the empirical risks, the output weights are then trained to minimize the output error and further balance and reduce the structural and the empirical risks of the SLFN. The performance of an SLFN-based classifier trained with the proposed scheme is evaluated in the simulation section in support of the proposed scheme.