TA3a.3

Achieving Extremely Low Communication Overhead in Federated Learning via Zeroth-Order SignSGD

Zhe Li, Dandan Liang, Rochester Institute of Technology, United States; Bicheng Ying, Google Inc., United States; Zidong Liu, ComboCurve Inc., United States; Rui Li, Haibo Yang, Rochester Institute of Technology, United States
Session:
TA3a: Recent Advances in Optimization for Machine Learning and Networking Lecture
Track:
Networks and Graphs
Location:
Scripps
Presentation Time:
Tue, 28 Oct, 09:05 - 09:30 PT (UTC -7)
Presentation
Discussion
Resources
No resources available.
Session TA3a
TA3a.1: Analog In-memory Training on General Non-ideal Resistive Elements: Understanding the Impact of Response Functions
Zhaoxian Wu, Quan Xiao, Rensselaer Polytechnic Institute, United States; Tayfun Gokmen, Omobayode Fagbohungbe, IBM T. J. Watson Research Center, United States; Tianyi Chen, Rensselaer Polytechnic Institute, United States
TA3a.2: Beyond Black-Box Analysis: Understanding the Role of Local Updates in Distributed Learning
Mingrui Liu, George Mason University, United States
TA3a.3: Achieving Extremely Low Communication Overhead in Federated Learning via Zeroth-Order SignSGD
Zhe Li, Dandan Liang, Rochester Institute of Technology, United States; Bicheng Ying, Google Inc., United States; Zidong Liu, ComboCurve Inc., United States; Rui Li, Haibo Yang, Rochester Institute of Technology, United States
TA3a.4: Revisiting Large-Scale Non-convex Distributionally Robust Optimization
QI Zhang, Arizona State University, United States; Yi Zhou, Texas A&M University, United States; Simon Khan, Ashley Prater-Bennette, Air Force Research Laboratory, United States; Lixin Shen, Syracuse University, United States; Shaofeng Zou, Arizona State University, United States
Contacts