Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mobility-Assisted Decentralized Federated Learning: Convergence Analysis and A Data-Driven Approach

Reza Jahani, Md Farhamdur Reza, Richeng Jin, Huaiyu Dai

TL;DR

This work tackles the performance degradation of decentralized federated learning (DFL) in sparse wireless networks with non-IID data by introducing mobility as a mechanism to improve information flow. It proves that mobility can enforce $B$-strong connectivity and derives a convergence bound for DFL under mobility, highlighting how the temporal evolution of the mixing matrix and data heterogeneity influence learning. To exploit this, it introduces two distribution-aware mobility strategies, Distribution-Aware Movement (DAM) and Distribution-Aware Cluster Center Movement (DCM), which guide mobile clients to visit locations with distinct data distributions to maximize information propagation. Empirical results on MNIST and CIFAR-10 show that DAM and DCM consistently outperform baselines, particularly under high heterogeneity, and generalize to larger networks and datasets, demonstrating the practical potential of mobility-enabled DFL in next-generation wireless networks. The findings offer a principled framework for deploying mobility to achieve near-uniform connectivity and mitigate non-IID effects in decentralized learning systems.

Abstract

Decentralized Federated Learning (DFL) has emerged as a privacy-preserving machine learning paradigm that enables collaborative training among users without relying on a central server. However, its performance often degrades significantly due to limited connectivity and data heterogeneity. As we move toward the next generation of wireless networks, mobility is increasingly embedded in many real-world applications. The user mobility, either natural or induced, enables clients to act as relays or bridges, thus enhancing information flow in sparse networks; however, its impact on DFL has been largely overlooked despite its potential. In this work, we systematically investigate the role of mobility in improving DFL performance. We first establish the convergence of DFL in sparse networks under user mobility and theoretically demonstrate that even random movement of a fraction of users can significantly boost performance. Building upon this insight, we propose a DFL framework that utilizes mobile users with induced mobility patterns, allowing them to exploit the knowledge of data distribution to determine their trajectories to enhance information propagation through the network. Through extensive experiments, we empirically confirm our theoretical findings, validate the superiority of our approach over baselines, and provide a comprehensive analysis of how various network parameters influence DFL performance in mobile networks.

Mobility-Assisted Decentralized Federated Learning: Convergence Analysis and A Data-Driven Approach

TL;DR

This work tackles the performance degradation of decentralized federated learning (DFL) in sparse wireless networks with non-IID data by introducing mobility as a mechanism to improve information flow. It proves that mobility can enforce -strong connectivity and derives a convergence bound for DFL under mobility, highlighting how the temporal evolution of the mixing matrix and data heterogeneity influence learning. To exploit this, it introduces two distribution-aware mobility strategies, Distribution-Aware Movement (DAM) and Distribution-Aware Cluster Center Movement (DCM), which guide mobile clients to visit locations with distinct data distributions to maximize information propagation. Empirical results on MNIST and CIFAR-10 show that DAM and DCM consistently outperform baselines, particularly under high heterogeneity, and generalize to larger networks and datasets, demonstrating the practical potential of mobility-enabled DFL in next-generation wireless networks. The findings offer a principled framework for deploying mobility to achieve near-uniform connectivity and mitigate non-IID effects in decentralized learning systems.

Abstract

Decentralized Federated Learning (DFL) has emerged as a privacy-preserving machine learning paradigm that enables collaborative training among users without relying on a central server. However, its performance often degrades significantly due to limited connectivity and data heterogeneity. As we move toward the next generation of wireless networks, mobility is increasingly embedded in many real-world applications. The user mobility, either natural or induced, enables clients to act as relays or bridges, thus enhancing information flow in sparse networks; however, its impact on DFL has been largely overlooked despite its potential. In this work, we systematically investigate the role of mobility in improving DFL performance. We first establish the convergence of DFL in sparse networks under user mobility and theoretically demonstrate that even random movement of a fraction of users can significantly boost performance. Building upon this insight, we propose a DFL framework that utilizes mobile users with induced mobility patterns, allowing them to exploit the knowledge of data distribution to determine their trajectories to enhance information propagation through the network. Through extensive experiments, we empirically confirm our theoretical findings, validate the superiority of our approach over baselines, and provide a comprehensive analysis of how various network parameters influence DFL performance in mobile networks.
Paper Structure (35 sections, 5 theorems, 68 equations, 8 figures, 2 tables, 5 algorithms)

This paper contains 35 sections, 5 theorems, 68 equations, 8 figures, 2 tables, 5 algorithms.

Table of Contents

  1. Introduction
  2. System Architecture
  3. Network Setup
  4. DFL System Model
  5. Baseline Mobility Models
  6. Static
  7. Random Movement
  8. Theoretical Analysis
  9. Mobility and B-Strong Connectivity
  10. Static
  11. Mobile Network
  12. Convergence Analysis
  13. Methodology intuition/motivation
  14. Distribution Aware Mobility-Assisted DFL
  15. Distribution Distance
  16. ...and 20 more sections

Key Result

Lemma 1

Let $\mathbf{X}^{(t)} = [ \mathbf{x}_1^{(t)},..., \mathbf{x}_C^{(t)} ] \in \mathbb{R}^{d \times N}$, $\mathbf{W}^{(t)}$ denote the doubly stochastic mixing matrix in round $t$, and $\mathbf{G}^{(t)} = [\mathbf{g}_1^{(t)},..., \mathbf{g}_C^{(t)}]$ where $\mathbf{g}_i^{(t)}=\nabla F_i (\mathbf{x}_i^{(

Figures (8)

  • Figure 1: Graphs representing the connectivity of clients in a network with a limited communication radius respectively in a static and mobile network
  • Figure 2: Test accuracy on MNIST dataset with 20 clients.
  • Figure 3: Robustness of the method in any configuration of $(|\mathcal{C}_m|$, $R_c$ and $R_m)$ using MNIST dataset with 20 clients.
  • Figure 4: $R_c$ threshold in a network with 20 clients.
  • Figure 5: $|\mathcal{C}_m|$ threshold in a DFL Network.
  • ...and 3 more figures

Theorems & Definitions (5)

  • Lemma 1: Stacked models representation
  • Corollary 1: Mixing Matrices Product
  • Theorem 1: Optimization error of global objective function
  • Lemma 2: Descent Lemma
  • Lemma 3: Consensus Control