面向边缘设备的基于模型融合剪枝的领域自适应联邦学习——DapperFL
摘要
联邦学习（FL）已成为边缘计算环境中一种突出的机器学习范式，使边缘设备能够在不共享私有数据的情况下协同优化全局模型。然而，现有FL框架因边缘计算中固有的系统异构性，尤其是在存在领域数据迁移的情况下，会导致效果下降。本文提出了一种异构FL框架DapperFL，以提高其在多个领域上的模型性能。在DapperFL中，我们引入了专用的模型融合剪枝（MFP）模块，为客户端生产个性化紧凑型本地模型，以解决系统异构性挑战。MFP模块通过融合来自本地和剩余领域的知识来修剪本地模型，确保对领域迁移具有鲁棒性。此外，我们设计了领域自适应正则化（DAR）模块进一步提高DapperFL的总体性能。DAR模块利用被修剪后的模型生成的正则化，以跨领域学习鲁棒表示。 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.
引用
@article{arxiv.2412.05823,
title = {DapperFL: Domain Adaptive Federated Learning with Model Fusion Pruning for Edge Devices},
author = {Yongzhe Jia and Xuyun Zhang and Hongsheng Hu and Kim-Kwang Raymond Choo and Lianyong Qi and Xiaolong Xu and Amin Beheshti and Wanchun Dou},
journal= {arXiv preprint arXiv:2412.05823},
year = {2024}
}
备注
Oral accepted by NeurIPS 2024