A Robust ELM Algorithm for Compensating the Effect of Node Fault and Weight Noise

Although the extreme learning machine (ELM) technique is an efficient and effective neural approach, there are still some downsides in the traditional ELM technique. When there are some outlier training samples, the trained neural network is usually with poor performance. Another issue is that when there are some noise and faults in the trained network, the performance of the trained network is also poor. This paper looks into the ELM technique under multiple imperfections, including outlier training samples, weight noise and node faults. This paper first identifies a regularization term for handling weight noise and node faults. To handle outlier training samples, the maximum correntropy criterion (MCC) concept is used in the objective function. A learning algorithm, namely, robust fault aware ELM algorithm (RFAELM), for faulty networks is then proposed. Simulation results show that the performance of the proposed algorithm is much better than that of two state-of-art robust algorithms.