Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Clustering-Based User Selection in Federated Learning: Metadata Exploitation for 3GPP Networks

Ce Zheng, Shiyao Ma, Ke Zhang, Chen Sun, Wenqi Zhang

TL;DR

The paper addresses realism gaps in federated learning by introducing an $HPPP$-based data partition model that captures data quantity heterogeneity and inter-user overlap, and a metadata-driven clustering strategy to minimize data correlation across clients. It leverages metadata such as user location to partition users into groups and selects one representative per group, reducing redundancy and increasing label diversity during training. Empirical evaluation on FMNIST and CIFAR-10 shows that clustering-based selection improves convergence stability and test accuracy in non-IID settings, particularly when the number of participating users per round is small, while maintaining comparable performance to baselines in IID scenarios. The work aligns with 3GPP standards by utilizing readily available location metadata and demonstrates practical benefits for large-scale cellular and edge-network deployments, with avenues for adaptive clustering and richer metadata in future work.

Abstract

Federated learning (FL) enables collaborative model training without sharing raw user data, but conventional simulations often rely on unrealistic data partitioning and current user selection methods ignore data correlation among users. To address these challenges, this paper proposes a metadatadriven FL framework. We first introduce a novel data partition model based on a homogeneous Poisson point process (HPPP), capturing both heterogeneity in data quantity and natural overlap among user datasets. Building on this model, we develop a clustering-based user selection strategy that leverages metadata, such as user location, to reduce data correlation and enhance label diversity across training rounds. Extensive experiments on FMNIST and CIFAR-10 demonstrate that the proposed framework improves model performance, stability, and convergence in non-IID scenarios, while maintaining comparable performance under IID settings. Furthermore, the method shows pronounced advantages when the number of selected users per round is small. These findings highlight the framework's potential for enhancing FL performance in realistic deployments and guiding future standardization.

Clustering-Based User Selection in Federated Learning: Metadata Exploitation for 3GPP Networks

TL;DR

The paper addresses realism gaps in federated learning by introducing an -based data partition model that captures data quantity heterogeneity and inter-user overlap, and a metadata-driven clustering strategy to minimize data correlation across clients. It leverages metadata such as user location to partition users into groups and selects one representative per group, reducing redundancy and increasing label diversity during training. Empirical evaluation on FMNIST and CIFAR-10 shows that clustering-based selection improves convergence stability and test accuracy in non-IID settings, particularly when the number of participating users per round is small, while maintaining comparable performance to baselines in IID scenarios. The work aligns with 3GPP standards by utilizing readily available location metadata and demonstrates practical benefits for large-scale cellular and edge-network deployments, with avenues for adaptive clustering and richer metadata in future work.

Abstract

Federated learning (FL) enables collaborative model training without sharing raw user data, but conventional simulations often rely on unrealistic data partitioning and current user selection methods ignore data correlation among users. To address these challenges, this paper proposes a metadatadriven FL framework. We first introduce a novel data partition model based on a homogeneous Poisson point process (HPPP), capturing both heterogeneity in data quantity and natural overlap among user datasets. Building on this model, we develop a clustering-based user selection strategy that leverages metadata, such as user location, to reduce data correlation and enhance label diversity across training rounds. Extensive experiments on FMNIST and CIFAR-10 demonstrate that the proposed framework improves model performance, stability, and convergence in non-IID scenarios, while maintaining comparable performance under IID settings. Furthermore, the method shows pronounced advantages when the number of selected users per round is small. These findings highlight the framework's potential for enhancing FL performance in realistic deployments and guiding future standardization.
Paper Structure (21 sections, 6 equations, 6 figures, 1 algorithm)

This paper contains 21 sections, 6 equations, 6 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model
  3. Federated learning
  4. User selection and broadcasting
  5. Local model updating
  6. Aggregation
  7. Data partition
  8. User Selection Strategy Based on Metadata
  9. User Clustering
  10. User Selection
  11. Simulation Results
  12. Experiment Setup
  13. IID
  14. Non-IID
  15. Simulation results
  16. ...and 6 more sections

Figures (6)

  • Figure 1: Illustration of data partition model where points in color (blue, red, green, pink) are data points generated by HPPP over$\mathcal{S}_U = [-(L+R)/2,(L+R)/2]^2$, and the black points are UEs uniformly distributed in $\mathcal{S}_U = [-L/2, L/2]^2$.
  • Figure 2: Non-IID data partition.
  • Figure 3: Performance evaluation of CIFAR-10 and FMNIST in IID scenario
  • Figure 4: Performance evaluation of CIFAR-10 and FMNIST in non-IID scenario
  • Figure 5: Impact of total user number $K$ on FL performance.
  • ...and 1 more figures