Kernel recursive maximum correntropy with variable center

In signal processing and machine learning, the maximum correntropy criterion with variable center (MCC-VC) has attracted more and more attention due to its robustness to non-zero mean noise. In this letter, we introduce MCC-VC into kernel space and develop the kernel recursive maximum correntropy with variable center (KRMCVC) algorithm. The proposed algorithm performs well in nonlinear signal processing, especially when data is disturbed by non-Gaussion and non-zero mean noises. In addition, we derive a quantized KRMCVC algorithm (QKRMCVC) based on the quantization method to control the network size. The superior performance of KRMCVC and QKRMCVC is confirmed in nonlinear system identification.