Broad Learning System Based on Maximum Correntropy Criterion

As an effective and efficient discriminative learning method, broad system (BLS) has received increasing attention due to its outstanding performance in various regression classification problems. However, the standard BLS is derived under minimum mean square error (MMSE) criterion, which is, of course, not always a good choice sensitivity outliers. To enhance robustness BLS, we propose this work adopt maximum correntropy criterion (MCC) train output weights, obtaining correntropy-based (C-BLS). Due inherent superiorities MCC, proposed C-BLS expected achieve excellent outliers while maintaining original Gaussian or noise-free environment. In addition, three alternative incremental algorithms, from weighted regularized least-squares solution rather than pseudoinverse formula, for are developed. With can be updated quickly without entire retraining process beginning when some new samples arrive network deems expanded. Experiments on data sets reported demonstrate desirable methods.