Saved in:
Bibliographic Details
Main Author: Yun, Vincent-Daniel
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.07142
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915716261740544
author Yun, Vincent-Daniel
author_facet Yun, Vincent-Daniel
contents Low-precision training has become crucial for reducing the computational and memory costs of large-scale deep learning. However, quantizing gradients introduces magnitude shrinkage, which can change how stochastic gradient descent (SGD) converges. In this study, we explore SGD convergence under a gradient shrinkage model, where each stochastic gradient is scaled by a factor \( q_k \in (0,1] \). We show that this shrinkage affect the usual stepsize \( μ_k \) with an effective stepsize \( μ_k q_k \), slowing convergence when \( q_{\min} < 1 \). With typical smoothness and bounded-variance assumptions, we prove that low-precision SGD still converges, but at a slower pace set by \( q_{\min} \), and with a higher steady error level due to quantization effects. We analyze theoretically how lower numerical precision slows training by treating it as gradient shrinkage within the standard SGD convergence setup.
format Preprint
id arxiv_https___arxiv_org_abs_2508_07142
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Why Does Stochastic Gradient Descent Slow Down in Low-Precision Training?
Yun, Vincent-Daniel
Machine Learning
Artificial Intelligence
Information Theory
Numerical Analysis
Low-precision training has become crucial for reducing the computational and memory costs of large-scale deep learning. However, quantizing gradients introduces magnitude shrinkage, which can change how stochastic gradient descent (SGD) converges. In this study, we explore SGD convergence under a gradient shrinkage model, where each stochastic gradient is scaled by a factor \( q_k \in (0,1] \). We show that this shrinkage affect the usual stepsize \( μ_k \) with an effective stepsize \( μ_k q_k \), slowing convergence when \( q_{\min} < 1 \). With typical smoothness and bounded-variance assumptions, we prove that low-precision SGD still converges, but at a slower pace set by \( q_{\min} \), and with a higher steady error level due to quantization effects. We analyze theoretically how lower numerical precision slows training by treating it as gradient shrinkage within the standard SGD convergence setup.
title Why Does Stochastic Gradient Descent Slow Down in Low-Precision Training?
topic Machine Learning
Artificial Intelligence
Information Theory
Numerical Analysis
url https://arxiv.org/abs/2508.07142