First-order methods with momentum such as Nesterov's fast gradient method are very useful for convex optimization problems, but can exhibit undesirable oscillations yielding slow convergence for some applications. An adaptive restarting scheme can improve the ... ...

请注册登录后继续浏览