A Robust and Regularized Extreme Learning Machine

In a moment when the study of outlier robustness within Extreme Learning Machine is still in its infancy, we propose a method that combines maximization of the hidden layer’s information transmission, through Batch Intrinsic Plasticity (BIP), with robust estimation of the output weights. This method named R-ELM/BIP generates a reliable solution in the presence of corrupted data with a good generalization capability and small weight norms. Computer experiments were carried out with three regression problems using traditional ELM, ELM with BIP, ELM using Iteratively Reweighted Least Squares as estimation method (ROB-ELM) and our proposal (R-ELM/BIP).