Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MoCFL: Mobile Cluster Federated Learning Framework for Highly Dynamic Network

Kai Fang, Jiangtao Deng, Chengzu Dong, Usman Naseem, Tongcun Liu, Hailin Feng, Wei Wang

TL;DR

MoCFL tackles robustness in highly dynamic mobile cluster FL by combining an affinity-based client selection mechanism for local feature extractor sharing with a memory-augmented global classifier trained on fused historical and current feature representations via MMD. The framework decouples models into a feature extractor and a classifier, enabling personalized local updates and a globally generalized classifier that mitigates forgetting under non-IID distributions and frequent client churn. Empirical results on the UNSW-NB15 intrusion dataset show MoCFL achieving superior accuracy and stability in both static and dynamic scenarios, with competitive time overhead. This approach advances privacy-preserving, adaptive FL for highly dynamic networks with heterogeneous data distributions.

Abstract

Frequent fluctuations of client nodes in highly dynamic mobile clusters can lead to significant changes in feature space distribution and data drift, posing substantial challenges to the robustness of existing federated learning (FL) strategies. To address these issues, we proposed a mobile cluster federated learning framework (MoCFL). MoCFL enhances feature aggregation by introducing an affinity matrix that quantifies the similarity between local feature extractors from different clients, addressing dynamic data distribution changes caused by frequent client churn and topology changes. Additionally, MoCFL integrates historical and current feature information when training the global classifier, effectively mitigating the catastrophic forgetting problem frequently encountered in mobile scenarios. This synergistic combination ensures that MoCFL maintains high performance and stability in dynamically changing mobile environments. Experimental results on the UNSW-NB15 dataset show that MoCFL excels in dynamic environments, demonstrating superior robustness and accuracy while maintaining reasonable training costs.

MoCFL: Mobile Cluster Federated Learning Framework for Highly Dynamic Network

TL;DR

MoCFL tackles robustness in highly dynamic mobile cluster FL by combining an affinity-based client selection mechanism for local feature extractor sharing with a memory-augmented global classifier trained on fused historical and current feature representations via MMD. The framework decouples models into a feature extractor and a classifier, enabling personalized local updates and a globally generalized classifier that mitigates forgetting under non-IID distributions and frequent client churn. Empirical results on the UNSW-NB15 intrusion dataset show MoCFL achieving superior accuracy and stability in both static and dynamic scenarios, with competitive time overhead. This approach advances privacy-preserving, adaptive FL for highly dynamic networks with heterogeneous data distributions.

Abstract

Frequent fluctuations of client nodes in highly dynamic mobile clusters can lead to significant changes in feature space distribution and data drift, posing substantial challenges to the robustness of existing federated learning (FL) strategies. To address these issues, we proposed a mobile cluster federated learning framework (MoCFL). MoCFL enhances feature aggregation by introducing an affinity matrix that quantifies the similarity between local feature extractors from different clients, addressing dynamic data distribution changes caused by frequent client churn and topology changes. Additionally, MoCFL integrates historical and current feature information when training the global classifier, effectively mitigating the catastrophic forgetting problem frequently encountered in mobile scenarios. This synergistic combination ensures that MoCFL maintains high performance and stability in dynamically changing mobile environments. Experimental results on the UNSW-NB15 dataset show that MoCFL excels in dynamic environments, demonstrating superior robustness and accuracy while maintaining reasonable training costs.

Paper Structure

This paper contains 16 sections, 15 equations, 7 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Federated Learning in Static Scenarios
  4. Federated Learning in Dynamic Scenarios
  5. Methodology
  6. MoCFL Framework Overview
  7. Update of Local Feature Extractors
  8. Update of Global Classifier
  9. Experiments and Analysis
  10. Comparison of Experimental Results in Static Scenarios
  11. Comparison of Experimental Results in Dynamic Scenarios
  12. Time Overhead Analysis of Federated Learning Strategies
  13. Conclusion
  14. Acknowledgment
  15. Dataset Description and Preprocessing
  16. ...and 1 more sections

Figures (7)

  • Figure 1: FL in Dynamic Mobile Clusters
  • Figure 2: MoCFL System Framework
  • Figure 3: Performance of FL Strategies Under Varying Client Activity Rates in UNSW-NB15 Dataset (10 Clients)
  • Figure 4: Performance of FL Strategies Under Varying Client Activity Rates in USNW-NB15 Dataset (50 Clients)
  • Figure 5: Distribution of Data Quantities for Different Attack Types in the UNSW-NB15 Dataset
  • ...and 2 more figures