We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math.OC

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Mathematics > Optimization and Control

Title: A Ring Topology-based Optimization Approach for Federated Learning in D2D Wireless Networks

Abstract: Federated learning (FL) is an emerging technique aiming at improving communication efficiency in distributed networks, where many clients often request to transmit their calculated parameters to an FL server simultaneously. However, in wireless networks, the above mechanism may lead to prolonged transmission time due to unreliable wireless transmission and limited bandwidth. This paper proposes an optimization approach to minimize the uplink transmission time for FL in wireless networks. The proposed approach consists of two major elements, namely a Ring All-reduce architecture that integrates D2D wireless communications to facilitate the FL process and a modified Ant Colony Optimization algorithm to identify the optimal composition of the Ring All-reduce architecture. Numerical results show that our proposed approach can significantly reduce the transmission time compared to traditional FL approaches. Notably, in scenarios applicable to large-scale transmissions where user devices are densely distributed, the reduction in transmission time compared to baseline policies can be over 50%.
Subjects: Optimization and Control (math.OC)
Cite as: arXiv:2212.02830 [math.OC]
  (or arXiv:2212.02830v1 [math.OC] for this version)

Submission history

From: Wei Tian [view email]
[v1] Tue, 6 Dec 2022 09:02:21 GMT (3825kb,D)

Link back to: arXiv, form interface, contact.