FederatedMesh: Collaborative Federated Learning for Medical Data Sharing in Mesh Networks

Edge computing is a paradigm that involves performing local processing on lightweight devices at the edge of networks to improve response times and reduce bandwidth consumption. While machine learning (ML) models can run on smaller computing devices at the edge, training ML models presents challenges for low-capacity devices. This paper aimed to evaluate the performance of Federated Learning (FL) - a distributed ML framework, when training a medical dataset using Raspberry Pi devices as client nodes. The testing accuracy, CPU usage, RAM memory usage and network performance were measured for different number of clients and epochs. The results showed that increasing the number of devices generally improved the testing accuracy, with the greatest improvement observed in the earlier epochs. However, increasing the number of devices also increased the CPU usage, with a significant increase observed in the later epochs. Additionally, the RAM memory usage increased slightly as the number of clients and epochs increased. The findings suggest that FL can be an effective way to train medical models using distributed devices, but careful consideration must be given to the trade-off between accuracy and computational resources.