Federated Learning Model Training for A Healthcare Domain

Abstract

In contrast to centralized data collection and model training, federated learning is a relatively new type of learning that does not involve centralized data collection. It is common in traditional machine learning pipelines to collect data from a variety of sources (such as mobile devices) and store it at a central location (such as a data center). A single machine learning model is trained on all of the data once it has been collected in the center. Because the data used to build and train the model must be transferred from the user's device to a central device, this approach is called "centralized learning". There are over 5 billion users of his mobile devices around the world. A large amount of data is generated by these users as a result of the use of cameras, microphones, and other sensors, such as accelerometers. This data can be used to build intelligent applications. In order to train machine/deep learning models and build intelligent applications, this data is collected in data centers.