Federated learning is a burgeoning and promising machine learning technique that employs distributed analytics and training of device-edge learning models. It is particularly well-suited for real-time and online learning at the edge with little or no reliance on the cloud for the transfer of learnable models, access to which could be costly and time-consuming, carrying with it severe consequences for latency-sensitive time-critical applications. Federated learning uplifts the use case of Internet of Things-based smart, intelligent, and autonomous systems by embedding machine and deep learning capabilities at the edge. It is also a natural solution for web-scale machine learning with a focus on user privacy and security, especially in industries such as healthcare and finance that are at risk of suffering massively painful consequences if users, devices, or edges are compromised. Moreover, federated learning has many advantages, such as location-based model customization, which is beneficial if the underlying data distributions differ among edge training clients but share a commonality, particularly in the case of large models, which translates into high training and inference costs, such as multiparty speech or natural language processing models. Such diverse, location-based models could also mitigate or reduce inference distortion and delay from not accurately modeling a user edge's specific request, since supposing 100,000 users are located in a given geographic area, one specific user may be requesting a different natural language response than another. The goal of this chapter is to help bridge the gap between theory and practical deployments of federated learning systems by exploring, analyzing, and addressing the computational overhead in federated learning models in the healthcare domain and providing several future directions. In Sect. 2, we outline and discuss just a few of the many use cases of federated learning in healthcare, including healthcare-related natural language processing applications.