[Paper Review] Joint Client Assignment and UAV Route Planning for Indirect-Communication Federated Learning

Jieming Bian, Cong Shen+1 more|arXiv (Cornell University)|Apr 21, 2023

Privacy-Preserving Technologies in Data4 citations

TL;DR

This paper proposes FedEx, a federated learning framework using UAVs as mobile intermediaries to enable indirect communication between clients and servers in areas without infrastructure. By jointly optimizing client assignment and UAV routing, FedEx ensures convergence under non-i.i.d. data and energy constraints, with experimental validation showing faster convergence and robustness across varying network conditions and energy budgets.

ABSTRACT

Federated Learning (FL) is a machine learning approach that enables the creation of shared models for powerful applications while allowing data to remain on devices. This approach provides benefits such as improved data privacy, security, and reduced latency. However, in some systems, direct communication between clients and servers may not be possible, such as remote areas without proper communication infrastructure. To overcome this challenge, a new framework called FedEx (Federated Learning via Model Express Delivery) is proposed. This framework employs mobile transporters, such as UAVs, to establish indirect communication channels between the server and clients. These transporters act as intermediaries and allow for model information exchange. The use of indirect communication presents new challenges for convergence analysis and optimization, as the delay introduced by the transporters' movement creates issues for both global model dissemination and local model collection. To address this, two algorithms, FedEx-Sync and FedEx-Async, are proposed for synchronized and asynchronized learning at the transporter level. Additionally, a bi-level optimization algorithm is proposed to solve the joint client assignment and route planning problem. Experimental validation using two public datasets in a simulated network demonstrates consistent results with the theory, proving the efficacy of FedEx.

Motivation & Objective

• Address the challenge of federated learning in remote, infrastructure-lacking environments where direct client-server communication is impossible.
• Overcome the convergence challenges introduced by indirect communication via mobile transporters, such as UAVs, which introduce variable delays.
• Develop a joint optimization framework for client assignment and UAV route planning that accounts for mobility, energy constraints, and communication delays.
• Ensure convergence of the FL process under non-convex loss functions and non-i.i.d. data distributions in both synchronized and asynchronous learning modes.
• Provide a systematic solution for deploying UAV-assisted federated learning in real-world, resource-constrained scenarios such as environmental monitoring or rural IoT.

Proposed method

• Propose a novel FL framework, FedEx, where UAVs act as mobile intermediaries to collect local models from clients and disseminate global models.
• Design two algorithms—FedEx-Sync and FedEx-Async—to manage model exchange timing at the UAV level, supporting both synchronized and asynchronous learning.
• Formulate a bi-level optimization problem to jointly solve client assignment and UAV route planning, minimizing total training time under energy and mobility constraints.
• Introduce a convergence analysis for both FedEx-Sync and FedEx-Async under non-convex, non-i.i.d. settings, proving convergence with explicit bounds.
• Incorporate moving speed and energy budget constraints into the optimization model, ensuring feasible and energy-efficient UAV trajectories.
• Use simulated environments with public datasets (FMNIST, SVHN) to evaluate performance under various non-i.i.d. data distributions and route designs.

Experimental results

Research questions

• RQ1 Can federated learning converge when communication between clients and server is indirect and delayed due to UAV mobility?
• RQ2 How do different UAV route planning strategies (e.g., SWS, Min-Max, Shortest-Total) affect convergence speed and model accuracy?
• RQ3 What is the impact of UAV number, speed, and energy budget on the convergence and performance of the FL system?
• RQ4 How does asynchrony in model exchange (FedEx-Async) compare to synchronization (FedEx-Sync) under varying mobility and workload conditions?
• RQ5 How does data non-i.i.d. distribution affect the performance of the joint client assignment and route planning strategy?

Key findings

• FedEx-Sync and FedEx-Async both converge under non-convex loss functions and non-i.i.d. data, with theoretical guarantees provided in the paper.
• The SWS (Shortest Worst-Case) route design achieves the best convergence performance in FedEx-Async, consistent with theoretical analysis.
• With fewer transporters (K=2), both FedEx-Sync and FedEx-Async converge, but convergence is slower due to longer RTTs and higher client load per UAV.
• Increasing the number of UAVs from 2 to 4 significantly improves convergence speed due to reduced per-UAV workload and shorter RTTs.
• Lower energy budgets (12 KJ vs. 15 KJ) degrade convergence speed in FedEx-Async due to sub-optimal cost function reach, while FedEx-Sync remains stable.
• Under low energy budgets, UAVs distribute work more evenly, leading to similar energy consumption and sub-optimal performance in the cost function.