Abstract
In this paper, we analyze robustness of adaptive gradient algorithms against continuous-time disturbance signals. It is shown that a kind of H∞ property holds for the adaptive gradient algorithms in the case where continuous-time outputs can be observed while it might not hold in the case where only sampled outputs can be observed. We also show an implementable method using the output over-sampling technique which gives better performance than the conventional and standard adaptive gradient algorithms. A numerical example is given to illustrate the effectiveness of the proposed method.
