Open Problem: Anytime Convergence Rate of Gradient Descent

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kornowski, Guy, Shamir, Ohad
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929392185245696
author Kornowski, Guy
Shamir, Ohad
author_facet Kornowski, Guy
Shamir, Ohad
contents Recent results show that vanilla gradient descent can be accelerated for smooth convex objectives, merely by changing the stepsize sequence. We show that this can lead to surprisingly large errors indefinitely, and therefore ask: Is there any stepsize schedule for gradient descent that accelerates the classic $\mathcal{O}(1/T)$ convergence rate, at \emph{any} stopping time $T$?
format Preprint
id arxiv_https___arxiv_org_abs_2406_13888
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Open Problem: Anytime Convergence Rate of Gradient Descent
Kornowski, Guy
Shamir, Ohad
Optimization and Control
Machine Learning
Recent results show that vanilla gradient descent can be accelerated for smooth convex objectives, merely by changing the stepsize sequence. We show that this can lead to surprisingly large errors indefinitely, and therefore ask: Is there any stepsize schedule for gradient descent that accelerates the classic $\mathcal{O}(1/T)$ convergence rate, at \emph{any} stopping time $T$?
title Open Problem: Anytime Convergence Rate of Gradient Descent
topic Optimization and Control
Machine Learning
url https://arxiv.org/abs/2406.13888