FISTA restart using an automatic estimation of the growth parameter

Jean-François Aujol (1) , Charles H Dossal (2) , Hippolyte Labarrière (2) , Aude Rondepierre (2)

1 IMB - Institut de Mathématiques de Bordeaux
2 INSA Toulouse - Institut National des Sciences Appliquées - Toulouse

In this paper, we propose a restart scheme for FISTA (Fast Iterative Shrinking-Threshold Algorithm). This method which is a generalization of Nesterov's accelerated gradient algorithm is widely used in the field of large convex optimization problems and it provides fast convergence results under a strong convexity assumption. These convergence rates can be extended for weaker hypotheses such as the \L{}ojasiewicz property but it requires prior knowledge on the function of interest. In particular, most of the schemes providing a fast convergence for non-strongly convex functions satisfying a quadratic growth condition involve the growth parameter which is generally not known. Recent works show that restarting FISTA could ensure a fast convergence for this class of functions without requiring any knowledge on the growth parameter. We improve these restart schemes by providing a better asymptotical convergence rate and by requiring a lower computation cost. We present numerical results emphasizing the efficiency of this method.

Domaines: Optimisation et contrôle [math.OC]

Jean-François Aujol, Charles H Dossal, Hippolyte Labarrière, Aude Rondepierre. FISTA restart using an automatic estimation of the growth parameter. 2022. ⟨hal-03153525v4⟩