Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DapperFL: Domain Adaptive Federated Learning with Model Fusion Pruning for Edge Devices

Yongzhe Jia, Xuyun Zhang, Hongsheng Hu, Kim-Kwang Raymond Choo, Lianyong Qi, Xiaolong Xu, Amin Beheshti, Wanchun Dou

TL;DR

DapperFL addresses the dual challenges of system heterogeneity and domain shifts in edge federated learning by introducing Model Fusion Pruning (MFP) to generate personalized compact local models and Domain Adaptive Regularization (DAR) to encourage cross-domain robust representations. A dedicated heterogeneous aggregation strategy preserves both local and global knowledge when combining pruned updates from diverse clients. Experimental results on digits and office-caltech benchmarks with 10 heterogeneous clients show DapperFL achieving up to 2.28% higher global accuracy than strong baselines and reducing local model footprints by 20–80%, demonstrating practical efficiency and robustness in multi-domain FL. Overall, the work offers a principled approach to scalable, domain-aware FL on resource-constrained edge devices, with potential impact on real-world deployments requiring both performance and efficiency across diverse environments.

Abstract

Federated learning (FL) has emerged as a prominent machine learning paradigm in edge computing environments, enabling edge devices to collaboratively optimize a global model without sharing their private data. However, existing FL frameworks suffer from efficacy deterioration due to the system heterogeneity inherent in edge computing, especially in the presence of domain shifts across local data. In this paper, we propose a heterogeneous FL framework DapperFL, to enhance model performance across multiple domains. In DapperFL, we introduce a dedicated Model Fusion Pruning (MFP) module to produce personalized compact local models for clients to address the system heterogeneity challenges. The MFP module prunes local models with fused knowledge obtained from both local and remaining domains, ensuring robustness to domain shifts. Additionally, we design a Domain Adaptive Regularization (DAR) module to further improve the overall performance of DapperFL. The DAR module employs regularization generated by the pruned model, aiming to learn robust representations across domains. Furthermore, we introduce a specific aggregation algorithm for aggregating heterogeneous local models with tailored architectures and weights. We implement DapperFL on a realworld FL platform with heterogeneous clients. Experimental results on benchmark datasets with multiple domains demonstrate that DapperFL outperforms several state-of-the-art FL frameworks by up to 2.28%, while significantly achieving model volume reductions ranging from 20% to 80%. Our code is available at: https://github.com/jyzgh/DapperFL.

DapperFL: Domain Adaptive Federated Learning with Model Fusion Pruning for Edge Devices

TL;DR

DapperFL addresses the dual challenges of system heterogeneity and domain shifts in edge federated learning by introducing Model Fusion Pruning (MFP) to generate personalized compact local models and Domain Adaptive Regularization (DAR) to encourage cross-domain robust representations. A dedicated heterogeneous aggregation strategy preserves both local and global knowledge when combining pruned updates from diverse clients. Experimental results on digits and office-caltech benchmarks with 10 heterogeneous clients show DapperFL achieving up to 2.28% higher global accuracy than strong baselines and reducing local model footprints by 20–80%, demonstrating practical efficiency and robustness in multi-domain FL. Overall, the work offers a principled approach to scalable, domain-aware FL on resource-constrained edge devices, with potential impact on real-world deployments requiring both performance and efficiency across diverse environments.

Abstract

Federated learning (FL) has emerged as a prominent machine learning paradigm in edge computing environments, enabling edge devices to collaboratively optimize a global model without sharing their private data. However, existing FL frameworks suffer from efficacy deterioration due to the system heterogeneity inherent in edge computing, especially in the presence of domain shifts across local data. In this paper, we propose a heterogeneous FL framework DapperFL, to enhance model performance across multiple domains. In DapperFL, we introduce a dedicated Model Fusion Pruning (MFP) module to produce personalized compact local models for clients to address the system heterogeneity challenges. The MFP module prunes local models with fused knowledge obtained from both local and remaining domains, ensuring robustness to domain shifts. Additionally, we design a Domain Adaptive Regularization (DAR) module to further improve the overall performance of DapperFL. The DAR module employs regularization generated by the pruned model, aiming to learn robust representations across domains. Furthermore, we introduce a specific aggregation algorithm for aggregating heterogeneous local models with tailored architectures and weights. We implement DapperFL on a realworld FL platform with heterogeneous clients. Experimental results on benchmark datasets with multiple domains demonstrate that DapperFL outperforms several state-of-the-art FL frameworks by up to 2.28%, while significantly achieving model volume reductions ranging from 20% to 80%. Our code is available at: https://github.com/jyzgh/DapperFL.

Paper Structure

This paper contains 17 sections, 7 equations, 7 figures, 6 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. DapperFL Design
  4. Overview of DapperFL
  5. Model Fusion Pruning on Edge Devices
  6. Domain Adaptive Regularization
  7. Heterogeneous Model Aggregation
  8. Experimental Evaluation
  9. Experimental Setup
  10. Performance Comparison
  11. Ablation Study
  12. Conclusion
  13. Introduction to Comparison Frameworks
  14. Default Hyper-parameters
  15. Learning Curves
  16. ...and 2 more sections

Figures (7)

  • Figure 1: A motivational example of DapperFL with heterogeneous devices and multiple domains.
  • Figure 2: Overview of DapperFL framework with two clients for each communication round.
  • Figure 3: Comparison of model accuracy of FedMP, NeFL, and DapperFL with different pruning ratios on the Digits and Office Caltech.
  • Figure 4: The effect of hyper-parameters in the MFP and DAR modules on model accuracy. "Office" in the legend represents the Office Caltech benchmark, and the average accuracy calculated for both benchmarks is labeled as "AVG".
  • Figure 5: Learning curves of global accuracies for FedAvg, FedDrop, FedProx, MOON, FedMP, FedSR, NeFL, FPL, and DapperFL on the Digits benchmark and the Office Caltech benchmark.
  • ...and 2 more figures