Analysis of Federated Learning as a Distributed Solution for Learning on Edge Devices
Saba F. Lameh, Wade Noble, Yasaman Amannejad, Arash Afshar · 2020
Sensors and smart devices are continuously collecting massive amounts of data. Today's state-of-the-art machine learning (ML) techniques, such as deep networks, are typically trained using cloud platforms, leveraging the elastic scalability of the cloud. For such processing, data from various sources need to be transferred to a cloud server. While this works well for some application domains, it is not suitable for all due to privacy and network overhead concerns. Sharing life logging photos and videos from cell phones and wearable devices can cause privacy concerns for users. To respond to the needs of such applications, federated learning (FL) is proposed as a distributed ML solution for learning on edge devices, such as cellphones. In FL, data does not leave users' devices. All users collaboratively train a model without sharing their data. Each user trains a local model with their data and shares the model updates with a FL server to aggregate and build a global model. Such training respects users' privacy while reducing the network overhead for transferring data to a central server. In this paper, we aim to study the FL approach for learning on edge devices. We investigate the performance of the FL model with various parameter settings and compare its accuracy and training time to a traditional learning technique that is trained on a central server. For this purpose, we train a convolutional deep learning network for image recognition in traditional and federated format. The results confirm that it is possible to conduct a complex learning task on distributed devices without sharing their data with a central server, and such federated models can achieve comparable accuracy to central models.