Broad learning system via adaptive maximum weighted correntropy

Broad Learning System (BLS) is widely used in various regression problems due to its simple structure and strong generalization ability. The standard optimized method for BLS is sensitive to the noise and outliers since it uses the Minimum Mean Square Error (MMSE) criterion, which may decrease the model's accuracy. As a solution, an Adaptive Maximum Weighted Correntropy - based BLS (AMWC-BLS) is proposed in this paper. Firstly, an adaptive maximum weighted correntropy criterion is presented to improve the performance and generalization ability of the model. Then, the AMWC-BLS is establised by embedding the AMWC criterion into the BLS. The proposed AMWC-BLS model can adjust its output weights facing the distinct characteristics of the input data and optimizing local data features. Hence, the AMWC-BLS model is able to better withstand the effects of outliers and noise and improve the robustness. Finally, the robustness and effectiveness of AMWC-BLS are demonstrated through the experiments on regression datasets.