Robust Multi-kernel Generalized Maximum Correntropy Filters

The traditional multi-kernel filtering algorithms are proposed by combining multiple kernel with the second similarity measure. Nevertheless, these multi-kernel filters suffer from both heavy computation burden and divergence for impulsive noise environments. To deal with these challenges for online applications, the multi-kernel generalized maximum correntropy (MKGMC) algorithm is developed for improving both precision and stability under the generalized maximum correntropy criterion. To alleviate the complex network structure of MKGMC, using the vector quantization strategy in MKGMC develops quantized MKGMC (QMKGMC) to reduce the computational and spatial complexity. From the viewpoints of precision and stability, the proposed algorithms outperform other typical compared algorithms on time series prediction.