1: \begin{abstract}

2: In the last decade, a considerable research effort has been devoted

3: to developing adaptive algorithms based on kernel functions. One of

4: the main features of these algorithms is that they form a family of

5: universal approximation techniques, solving problems with

6: nonlinearities elegantly. In this paper, we present data-selective

7: adaptive {kernel normalized least-mean square (KNLMS)} algorithms

8: that can increase their learning rate and reduce their computational

9: complexity. In fact, these methods deal with kernel expansions,

10: creating a growing structure also known as the dictionary, whose

11: size depends on the number of observations and their innovation.

12: {The algorithms described herein use an adaptive step-size to

13: accelerate the learning and can offer an excellent tradeoff between

14: convergence speed and steady state, which allows them to solve

15: nonlinear filtering and estimation problems with a large number of

16: parameters without requiring a large computational cost.} The

17: data-selective update scheme also limits the number of operations

18: performed and the size of the dictionary created by the kernel

19: expansion, saving computational resources and dealing with one of

20: the major problems of kernel adaptive algorithms. A statistical

21: analysis is carried out along with a computational complexity

22: analysis of the proposed algorithms.

23:  {Simulations show that the proposed KNLMS algorithms

24: outperform existing algorithms in examples of nonlinear system

25: identification and prediction of a time series originating from a

26: nonlinear difference equation.}

27: \end{abstract}

28: