Gradient-free algorithm for saddle point problems under overparametrization

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Statkevich, Ekaterina, Bondar, Sofiya, Dvinskikh, Darina, Gasnikov, Alexander, Lobanov, Aleksandr
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911903458000896
author Statkevich, Ekaterina
Bondar, Sofiya
Dvinskikh, Darina
Gasnikov, Alexander
Lobanov, Aleksandr
author_facet Statkevich, Ekaterina
Bondar, Sofiya
Dvinskikh, Darina
Gasnikov, Alexander
Lobanov, Aleksandr
contents This paper focuses on solving a stochastic saddle point problem (SPP) under an overparameterized regime for the case, when the gradient computation is impractical. As an intermediate step, we generalize Same-sample Stochastic Extra-gradient algorithm (Gorbunov et al., 2022) to a biased oracle and estimate novel convergence rates. As the result of the paper we introduce an algorithm, which uses gradient approximation instead of a gradient oracle. We also conduct an analysis to find the maximum admissible level of adversarial noise and the optimal number of iterations at which our algorithm can guarantee achieving the desired accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2406_02308
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Gradient-free algorithm for saddle point problems under overparametrization
Statkevich, Ekaterina
Bondar, Sofiya
Dvinskikh, Darina
Gasnikov, Alexander
Lobanov, Aleksandr
Optimization and Control
This paper focuses on solving a stochastic saddle point problem (SPP) under an overparameterized regime for the case, when the gradient computation is impractical. As an intermediate step, we generalize Same-sample Stochastic Extra-gradient algorithm (Gorbunov et al., 2022) to a biased oracle and estimate novel convergence rates. As the result of the paper we introduce an algorithm, which uses gradient approximation instead of a gradient oracle. We also conduct an analysis to find the maximum admissible level of adversarial noise and the optimal number of iterations at which our algorithm can guarantee achieving the desired accuracy.
title Gradient-free algorithm for saddle point problems under overparametrization
topic Optimization and Control
url https://arxiv.org/abs/2406.02308