复杂网络上的加速核学习算法OA

In this paper,the kernel learning problem over complex networks is studied,and the non-linear coupling functions between nodes were learned by using input-output data pairs.A kernel least mean square algorithm was designed,using the Random Fourier Feature method to reduce the compu-tational and storage complexity.To address the problem of gradient noise caused by the stochastic gradient descent algorithm,the kernel reproducing gradient descent algorithm was used to solve the mean square error function,which reduces the variance of gradient noise.Moreover,the adaptive mo-mentum strategy was adopted to accelerate the convergence of the kernel least mean square algorithm.Theoretical analyses are provided to show that the accelerated kernel learning algorithm is convergent if the learning rate meets certain conditions.Simulation examples verify the superiority of the proposed algorithm in terms of convergence speed and time cost.