Robust Deep Stochastic Configuration Network Modeling Method Based on Kernel Density Estimation

In order to alleviate the negative impact of noise on the accuracy of deep stochastic configuration network modeling, a robust deep stochastic configuration network modeling method based on kernel density estimation is proposed. The output weights of each hidden layer of deep stochastic configuration networks are obtained by solving the weighted least square problem, where the kernel density estimation method is employed to set the penalty weights of training samples. In addition, the alternating optimization technique is applied to update the penalty weights and hidden layer output weights. The effectiveness of the proposed method is tested and evaluated by using function approximation and the benchmark dataset. The results show that the proposed method can effectively alleviate the impact of noise on modeling accuracy, which is valuable for applications in the field of robust modeling.