Advancing the lower bounds: An accelerated, stochastic, second-order method with optimal adaptation to inexactness

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Agafonov, Artem, Kamzolov, Dmitry, Gasnikov, Alexander, Kavis, Ali, Antonakopoulos, Kimon, Cevher, Volkan, Takáč, Martin
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910459440922624
author Agafonov, Artem
Kamzolov, Dmitry
Gasnikov, Alexander
Kavis, Ali
Antonakopoulos, Kimon
Cevher, Volkan
Takáč, Martin
author_facet Agafonov, Artem
Kamzolov, Dmitry
Gasnikov, Alexander
Kavis, Ali
Antonakopoulos, Kimon
Cevher, Volkan
Takáč, Martin
contents We present a new accelerated stochastic second-order method that is robust to both gradient and Hessian inexactness, which occurs typically in machine learning. We establish theoretical lower bounds and prove that our algorithm achieves optimal convergence in both gradient and Hessian inexactness in this key setting. We further introduce a tensor generalization for stochastic higher-order derivatives. When the oracles are non-stochastic, the proposed tensor algorithm matches the global convergence of Nesterov Accelerated Tensor method. Both algorithms allow for approximate solutions of their auxiliary subproblems with verifiable conditions on the accuracy of the solution.
format Preprint
id arxiv_https___arxiv_org_abs_2309_01570
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Advancing the lower bounds: An accelerated, stochastic, second-order method with optimal adaptation to inexactness
Agafonov, Artem
Kamzolov, Dmitry
Gasnikov, Alexander
Kavis, Ali
Antonakopoulos, Kimon
Cevher, Volkan
Takáč, Martin
Optimization and Control
We present a new accelerated stochastic second-order method that is robust to both gradient and Hessian inexactness, which occurs typically in machine learning. We establish theoretical lower bounds and prove that our algorithm achieves optimal convergence in both gradient and Hessian inexactness in this key setting. We further introduce a tensor generalization for stochastic higher-order derivatives. When the oracles are non-stochastic, the proposed tensor algorithm matches the global convergence of Nesterov Accelerated Tensor method. Both algorithms allow for approximate solutions of their auxiliary subproblems with verifiable conditions on the accuracy of the solution.
title Advancing the lower bounds: An accelerated, stochastic, second-order method with optimal adaptation to inexactness
topic Optimization and Control
url https://arxiv.org/abs/2309.01570