Kernel incremental meta-learning with processed feedback

Abstract

In this letter, a recurrent kernel online learning algorithm with a processed feedback is proposed. The delayed output is processed by a well designed nonlinear piecewise function, which strengthens or weakens the feedback in response to different situation. Furthermore, the algorithm includes a data-dependent adaptive learning rate which evolves from kernel incremental meta-learning algorithm. Experimental results show that the novel algorithm outperforms both the kernel adaptive filter with multiple feedback and the kernel algorithm with single feedback in terms of convergence speed and estimation accuracy.