From differential equation solvers to accelerated first-order methods for convex optimization

报告题目:From differential equation solvers to accelerated first-order methods for convex optimization


邀请人:黄学海

报告人:陈龙  教授  美国加州大学尔湾分校

时间:2019年7月18日(周四)14:30-15:30

地点:红瓦楼723

摘要:Convergence analysis of accelerated first-order methods for convex optimization problems are presented  from the point of view of ordinary differential  equation (ODE) solvers. Two resolution ODEs are  derived for accelerated gradient methods.  Numerical discretizations for these resolution  ODEs are considered and its convergence analyses are  established via tailored Lyapunov functions.  The ODE solvers approach can not only cover existing methods,  such as Nesterov's accelerated gradient method and FISTA, but also produce a large class of  new algorithms that possesses optimal convergence rates. This is a joint work with Hao Luo from Sichuan University