The rapidly spreading and smart mobile/IoT devices worldwide are generating a massive amount of data. This huge data has been both a motivation and a key enabler for Machine Learning (ML)-based intelligence. Cloud Computing has been leveraged to run Centralized Machine/Deep Learning (ML/DL) where the user’s data is offloaded to central clouds for ML/DL processing. This centralized approach produces accurate models and algorithms since model training is happening over data collected from a large number of end devices. The more personal data collected and processed at the cloud, the more accurate the algorithms and the models become. Hence, better algorithms come at the cost of privacy. Moreover, Centralized Machine/Deep Learning on Clouds has several issues especially when data is sensitive or expensive to centralize. Edge Learning has been introduced to overcome high-latency and increased network traffic issues of ML processing on clouds. It leverages Edge/Fog nodes which are placed 1- to 2-hops away from end mobile/IoT devices. It resulted in a faster model downloads to these devices and allowed applications to respond faster to events. However, privacy has remained an issue and several sectors, such as healthcare and banking, could not fully benefit from centralized ML/DL. Hence, in 2017, Google introduced Federated Learning, which allows creating models that are competitive to centralized ML/DL models, without compromising on user privacy. With Federated learning the raw data is kept in place and the training is moved to the local end devices. A number of clients (mobile/IoT nodes), orchestrated by a central server, train models on their local devices. The central server collects training data from end devices, aggregates this data, and updates a global model that is shared again with end devices for further training. This iterative process continues until the model reaches a high accuracy. Federated Learning (FL) is a great tool that can assist with health-related applications. Given smartphones widespread and their hardware capabilities, they can be leveraged to collect sensory data necessary to diagnose several diseases. Federated Learning is a great tool for mHealth applications to give recommendations or to alert patients and/or the medical staff about the likelihood of a disease or a health issue. It can complement the efforts of the medical system, specially when it is overloaded in scenarios such as the case for COVID-19 nowadays. The architecture of the underlying system for Federated Learning has relied on Cloud and/or Edge computing approaches for running the central server. (1) Cloud-based Federated Learning: In this approach, the central server runs at a cloud server. (2) Edge-based Federated Learning: was leveraged to overcome the communication issue of cloud-based FL, by running the central server on Edge/Fog nodes. (3) Hierarchical Federated Learning which leverages both Cloud and Edge and the model aggregation happens at these levels. This hierarchical approach covers more clients, compared to Edge-based FL, while at the same time limitts the communication overhead introduced by Cloud-based FL. Relying on Edge Computing, even while integrating it with Cloud Computing, is not sufficient. Edge Computing suffers from a high-cost deployment barrier. Moreover, it cannot be utilized in network-challenged environments or by clients who are disconnected from the Internet most of the day, such as “Blue-collar workers" in construction sites in Qatar. However, it is important to include as many clients as possible for health applications that are performing remote monitoring and need continuous connectivity to handle emergencies and perform diagnosis in real-time. To overcome these limitations, we propose a cheaper and more scalable systems approach which relies on FemtoClouds. The idea of FemtoClouds is to leverage the compute resources of mobile device pools that exist within close proximity (0-hops away), for a period of time. This could vary from very stable environments, like households and a classroom, to more semi-stable environments like coffee shop or trains, etc. FemtoClouds and Federated Learning have common challenges brought by the participating mobile/IoT devices. The devices running the learning and other analysis tasks are not dedicated, heterogeneous, and have many resources constraints. In this research, we will build the underlying FemtoClouds system and algorithms that can efficiently support Federated Learning in mHealth applications. Specifically, we will address two of the main challenges that are common in both Federated Learning and beyond-Edge FemtoClouds; resource awareness/management, and incentives. New solutions need to be created to handle these issues while taking the requirements and the constraints of the two fields into account. We will also build an mHealth application that leverages Federated Learning and integrate it with the FemtoClouds system.