Year: 2026
Type: article
Abstract: Stochastic optimization naturally appears in many application areas, including machine learning. Our goal is to go further in the analysis of the stochastic average gradient accelerated (SAGA) algorithm. To achieve this, we introduce a new [Formula: see text]-SAGA algorithm which interpolates between the stochastic gradient descent ([Formula: see text]) and the SAGA algorithm ([Formula: see text]). First, we investigate the almost sure convergence of this new algorithm with decreasing step which allows us to avoid the restrictive strong convexity and Lipschitz gradient hypotheses associated with the objective function. Second, we establish a central limit theorem for the [Formula: see text]-SAGA algorithm. Finally, we provide the nonasymptotic [Formula: see text] rates of convergence.
DOI: 10.1137/25m179748x
Authors: Bernard Bercu, Luis Fredes, Eméric Gbaguidi
Institutions: Université de Bordeaux, Institut de Mathématiques de Bordeaux
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: Agence Nationale de la Recherche
OpenAlex ID: W4403116298
