Communication Efficient Distributed Newton Method over Unreliable Networks

Authors

  • Ming Wen Fudan University
  • Chengchang Liu The Chinese University of Hong Kong
  • Yuedong Xu Fudan University

DOI:

https://doi.org/10.1609/aaai.v38i14.29513

Keywords:

ML: Distributed Machine Learning & Federated Learning, ML: Optimization

Abstract

Distributed optimization in resource constrained devices demands both communication efficiency and fast convergence rates. Newton-type methods are getting preferable due to their superior convergence rates compared to the first-order methods. In this paper, we study a new problem in regard to the second-order distributed optimization over unreliable networks. The working devices are power-limited or operate in unfavorable wireless channels, experiencing packet losses during their uplink transmission to the server. Our scenario is very common in real-world and leads to instability of classical distributed optimization methods especially the second-order methods because of their sensitivity to the imprecision of local Hessian matrices. To achieve robustness to high packet loss, communication efficiency and fast convergence rates, we propose a novel distributed second-order method, called RED-New (Packet loss Resilient Distributed Approximate Newton). Each iteration of RED-New comprises two rounds of light-weight and lossy transmissions, in which the server aggregates the local information with a new developed scaling strategy. We prove the linear-quadratic convergence rate of RED-New. Experimental results demonstrate its advantage over first-order and second-order baselines, and its tolerance to packet loss rate ranging from 5% to 40%.

Published

2024-03-24

How to Cite

Wen, M., Liu, C., & Xu, Y. (2024). Communication Efficient Distributed Newton Method over Unreliable Networks. Proceedings of the AAAI Conference on Artificial Intelligence, 38(14), 15832-15840. https://doi.org/10.1609/aaai.v38i14.29513

Issue

Section

AAAI Technical Track on Machine Learning V