15
15

A first-order primal-dual method with adaptivity to local smoothness

Abstract

We consider the problem of finding a saddle point for the convex-concave objective minxmaxyf(x)+Ax,yg(y)\min_x \max_y f(x) + \langle Ax, y\rangle - g^*(y), where ff is a convex function with locally Lipschitz gradient and gg is convex and possibly non-smooth. We propose an adaptive version of the Condat-V\~u algorithm, which alternates between primal gradient steps and dual proximal steps. The method achieves stepsize adaptivity through a simple rule involving A\|A\| and the norm of recently computed gradients of ff. Under standard assumptions, we prove an O(k1)\mathcal{O}(k^{-1}) ergodic convergence rate. Furthermore, when ff is also locally strongly convex and AA has full row rank we show that our method converges with a linear rate. Numerical experiments are provided for illustrating the practical performance of the algorithm.

View on arXiv
Comments on this paper

We use cookies and other tracking technologies to improve your browsing experience on our website, to show you personalized content and targeted ads, to analyze our website traffic, and to understand where our visitors are coming from. See our policy.