An Emulation Platform for Wireless Ad Hoc Federated Learning: Design, Implementation, and Case Study

Wireless Ad Hoc Federated Learning (WAFL) is a distributed machine learning paradigm. In WAFL, participant devices collaboratively train their local models through direct wireless communication when in contact with each other. However, previous research in WAFL has been predominantly simulation-based. Also, given the nascent nature of research in ad hoc federated learning, existing federated learning frameworks offer limited support for ad hoc applications. In consideration of the above, this paper presents the WAFL Testbed, an emulation platform for WAFL projects. In its current form, it consists of 10 participant devices and a controller (for reproducibility) over a local area network, with potential support for a variable number of devices and diverse interaction behaviors. Further, to demonstrate its functionality and the experiment workflow, this paper also presents the details from the deployment of the WAFL Multi-Layer Perceptron project in multiple configurations on the WAFL Testbed.