Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

pFedWN: A Personalized Federated Learning Framework for D2D Wireless Networks with Heterogeneous Data

Zhou Ni, Masoud Ghazikor, Morteza Hashemi

TL;DR

pFedWN tackles data heterogeneity in decentralized FL by introducing a server-free PFL framework for D2D wireless networks. It jointly optimizes neighbor selection conditioned on wireless channels and data similarity using an EM-based weight assignment, enabling personalized aggregation without a central server. The paper provides convergence guarantees for both strong convex and non-convex losses and demonstrates through simulations on CIFAR-10/100 and MNIST that pFedWN improves target-client performance under non-IID data and various channel conditions, while reducing communication overhead by limiting participating neighbors. The work enables robust, scalable personalization for wireless edge networks and suggests future extensions to handle broader heterogeneity and multi-objective optimization.

Abstract

Traditional Federated Learning (FL) approaches often struggle with data heterogeneity across clients, leading to suboptimal model performance for individual clients. To address this issue, Personalized Federated Learning (PFL) emerges as a solution to the challenges posed by non-independent and identically distributed (non-IID) and unbalanced data across clients. Furthermore, in most existing decentralized machine learning works, a perfect communication channel is considered for model parameter transmission between clients and servers. However, decentralized PFL over wireless links introduces new challenges, such as resource allocation and interference management. To overcome these challenges, we formulate a joint optimization problem that incorporates the underlying device-to-device (D2D) wireless channel conditions into a server-free PFL approach. The proposed method, dubbed pFedWN, optimizes the learning performance for each client while accounting for the variability in D2D wireless channels. To tackle the formulated problem, we divide it into two sub-problems: PFL neighbor selection and PFL weight assignment. The PFL neighbor selection is addressed through channel-aware neighbor selection within unlicensed spectrum bands such as ISM bands. Next, to assign PFL weights, we utilize the Expectation-Maximization (EM) method to evaluate the similarity between clients' data and obtain optimal weight distribution among the chosen PFL neighbors. Empirical results show that pFedWN provides efficient and personalized learning performance with non-IID and unbalanced datasets. Furthermore, it outperforms the existing FL and PFL methods in terms of learning efficacy and robustness, particularly under dynamic and unpredictable wireless channel conditions.

pFedWN: A Personalized Federated Learning Framework for D2D Wireless Networks with Heterogeneous Data

TL;DR

pFedWN tackles data heterogeneity in decentralized FL by introducing a server-free PFL framework for D2D wireless networks. It jointly optimizes neighbor selection conditioned on wireless channels and data similarity using an EM-based weight assignment, enabling personalized aggregation without a central server. The paper provides convergence guarantees for both strong convex and non-convex losses and demonstrates through simulations on CIFAR-10/100 and MNIST that pFedWN improves target-client performance under non-IID data and various channel conditions, while reducing communication overhead by limiting participating neighbors. The work enables robust, scalable personalization for wireless edge networks and suggests future extensions to handle broader heterogeneity and multi-objective optimization.

Abstract

Traditional Federated Learning (FL) approaches often struggle with data heterogeneity across clients, leading to suboptimal model performance for individual clients. To address this issue, Personalized Federated Learning (PFL) emerges as a solution to the challenges posed by non-independent and identically distributed (non-IID) and unbalanced data across clients. Furthermore, in most existing decentralized machine learning works, a perfect communication channel is considered for model parameter transmission between clients and servers. However, decentralized PFL over wireless links introduces new challenges, such as resource allocation and interference management. To overcome these challenges, we formulate a joint optimization problem that incorporates the underlying device-to-device (D2D) wireless channel conditions into a server-free PFL approach. The proposed method, dubbed pFedWN, optimizes the learning performance for each client while accounting for the variability in D2D wireless channels. To tackle the formulated problem, we divide it into two sub-problems: PFL neighbor selection and PFL weight assignment. The PFL neighbor selection is addressed through channel-aware neighbor selection within unlicensed spectrum bands such as ISM bands. Next, to assign PFL weights, we utilize the Expectation-Maximization (EM) method to evaluate the similarity between clients' data and obtain optimal weight distribution among the chosen PFL neighbors. Empirical results show that pFedWN provides efficient and personalized learning performance with non-IID and unbalanced datasets. Furthermore, it outperforms the existing FL and PFL methods in terms of learning efficacy and robustness, particularly under dynamic and unpredictable wireless channel conditions.
Paper Structure (23 sections, 12 theorems, 78 equations, 9 figures, 3 tables, 2 algorithms)

This paper contains 23 sections, 12 theorems, 78 equations, 9 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. System Model
  4. D2D Learning Model
  5. Channel-Aware PFL Neighbor Selection
  6. Problem Formulation and Proposed Solution
  7. PFL Problem Formulation
  8. Proposed PFL Weight Assignment Methodology
  9. Convergence Analysis
  10. Strong Convex Loss Function
  11. Convergence Analysis with Non-Convex Loss Function
  12. Numerical Evaluation
  13. Evaluations Setup
  14. Impacts of Channel Conditions on PFL Neighbor Selection
  15. Effectiveness of the EM Algorithm and PFL Results
  16. ...and 8 more sections

Key Result

Theorem 1

Under Assumptions 1-3, and given the local model update rule in Eq. eq:12 for the selected neighbors $m \in [\boldsymbol{M}_{n}]$, and assuming that the local models $\omega_{m}$ are bounded, the pFedWN algorithm converges at a rate of $\mathcal{O}(\gamma^T)$ if $\alpha^{2}(2-\alpha)(1-\eta\mu)^{E}

Figures (9)

  • Figure 1: Performance comparison of a typical target client vs. the global model performance (FedAvg) for FL with 11 clients with non-IID and unbalanced CIFAR-10 dataset.
  • Figure 2: In the proposed system model, the target client selects a set of neighbors based on the wireless channel conditions. Then, the target client assigns model aggregation weights.
  • Figure 3: Example time slot for clients in the network. CCM: Communication Channel Measurement; DSE: Data Similarity Estimation; MPS: Model Parameters Sharing.
  • Figure 4: PFL-selected neighbors (SNs) as a function of transmission error probabilities to the target client (TC) across three different cases.
  • Figure 5: Average number of selected neighbors as a function of the number of subchannels $|\boldsymbol{F}|$, SINR thresholds $\gamma_{th}$, and PPP network density.
  • ...and 4 more figures

Theorems & Definitions (20)

  • Theorem 1
  • proof
  • Lemma 1
  • Lemma 2
  • Lemma 3
  • Lemma 4
  • Theorem 2
  • proof
  • Theorem 1
  • proof
  • ...and 10 more