logo
CanyamAI summaries for academic research
language
MobileMobile
  • Phone FeatureWeChat Mini Program
EnglishEnglish
noticeNotice
BlogBlog
delete
arrow
Return

HAGC: A Hardware-Aware Gradient Compression framework for distributed deep learning

delete2026-03-09
delete0
PRE
AI
A
Aiqiang Yang *
J
jie Liu
B
Bo Yang
Z
Zhang, Xiang, Ph. D. Massachusetts Institute of Technology
Q
Qinglin Wang
Z
Zeyao Mo
K
Keqin Li
DOI：10.1016/j.sysarc.2026.103770delete
deleteOriginal
deleteOriginal request for help
deleteShare
deleteSave
Abstract

Abstract

En 中文
• Bilateral Hadamard Transforms shift workloads to Tensor Cores. • Error feedback and co-design ensure throughput and convergence stability. • Achieves up to 3.15x speedup and 2.9x energy reduction on A100 GPUs.
Keywords:
Hadamard Transform
Gradient Compression
Tensor Cores
Distributed Deep Learning
Energy Efficiency

Journal

Journal of Systems Architecture cover
Journal of Systems Architecture
IF:
4.1
Papers:
3.0K
Citations:
4.2K

Organization

N
national university of defense technology
Scholars:
5.0K
Papers: 1.5K
Citations: 0
SUNY New Paltz cover
SUNY New Paltz
Scholars:
169
Papers: 199
Citations: 324
C
China Academy of Engineering Physics
Scholars:
1.0K
Papers: 340
Citations: 2
researcher View more organizations
Cited Papers

Cited Papers

No cited papers available