ACSP-FL: Adaptive Client Selection with Personalization for Communication Efficient Federated Learning
Abstract: Federated Learning (FL) is a distributed approach to collaboratively training machine learning models. FL requires a high level of communication between the devices and a central server, thus imposing several challenges, including communication bottlenecks and network scalability. This article introduces ACSP-FL, a solution to reduce the overall communication and computation costs for training a model in FL environments. ACSP-FL employs a client selection strategy that dynamically adapts the number of devices training the model and the number of rounds required to achieve convergence. Moreover, ACSP-FL enables model personalization to improve clients performance. A use case based on human activity recognition datasets aims to show the impact and benefits of ACSP-FL when compared to state-of-the-art approaches. Experimental evaluations show that ACSP-FL minimizes the overall communication and computation overheads to train a model and converges the system efficiently. In particular, ACSP-FL reduces communication up to 95% compared to literature approaches while providing good convergence even in scenarios where data is distributed differently, non-independent and identical way between client devices.
The following datasets are available:
- CIFAR-10
- MNIST
- Motion sense
- UCI-HAR
- ExtraSensory
--clients-c: number of clients--model-m: model to be used in the simulation (e.g., DNN, CNN, or Logistic Regression)--client-selection-: client selection method--dataset-d: Dataset to be used by the clients--local-epochs-e: number of local epochs in each client--rounds-r: total number of communication rounds--poc-: percentagem of clients to be selected using PoC--decay-: decay factor for DEEV and ACSP-FL--personalization-: enables personalization and partial model sharing--shared-layers-: number of layers to be shared in the personalization phase
tip: The ACSP-FL solution is the DEEV + Personalization enabled
It is important to generate new images for both the Client and the Server with the Dockerfile from both directories. Then replace the image in the create_dockercompose.py script
python create_dockercompose.py --client-selection='DEEV' --dataset='MNIST'
--model='DNN' --epochs=1 --round=10 --clients=50 docker compose -f <compose-file.yaml> --compatibility up 
@article{acsp-fl,
title = {Adaptive client selection with personalization for communication efficient Federated Learning},
journal = {Ad Hoc Networks},
volume = {157},
pages = {103462},
year = {2024},
issn = {1570-8705},
doi = {https://doi.org/10.1016/j.adhoc.2024.103462},
url = {https://www.sciencedirect.com/science/article/pii/S1570870524000738},
author = {Allan M. {de Souza} and Filipe Maciel and Joahannes B.D. {da Costa} and Luiz F. Bittencourt and Eduardo Cerqueira and Antonio A.F. Loureiro and Leandro A. Villas},
}