[其他] ICML-Quasi-global Momentum Accelerating Decentralized Deep Learning on Hete ... [推广有奖]

1关注
167
粉丝

已卖：5640份资源

泰斗

还不是VIP/贵宾

威望: 0 级
论坛币: 13322 个
通用积分: 711.3950
学术水平: 459 点
热心指数: 484 点
信用等级: 414 点
经验: 115899 点
帖子: 4676
精华: 0
在线时间: 16723 小时
注册时间: 2013-1-2
最后登录: 2025-12-14

楼主

Mujahida

发表于 2025-7-27 15:51:22 |AI写论文

是否 +2 论坛币

k人参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

Quasi-Global Momentum:
   Accelerating Decentralized Deep Learning on Heterogeneous Data

         Tao Lin 1 Sai Praneeth Karimireddy 1 Sebastian U. Stich 1 Martin Jaggi 1

         Abstract             is kept locally (never transmitted during training).
Decentralized training of deep learning models    A key challenge—in particular in the second scenario—is
is a key element for enabling data privacy and    the large heterogeneity (non-i.i.d.-ness) in the data present
...