CACHING TECHNIQUES FOR REDUCING THE COMMUNICATION COST OF FEDERATED LEARNING IN IOT ENVIRONMENTS

Abstract

Federated Learning (FL) introduces a new way to use data for machine learning, emphasizing privacy by handling data locally at the network’s edge, instead of large, centralized servers. This thesis aims to improve FL on the Internet of Things (IoT) settings by developing and applying sophisticated caching methods. These methods are intended to lower the communication demands that come with regularly updating the model between the network's edge devices and the server. High communication costs can limit the use of FL, especially in situations involving many IoT devices with restricted bandwidth and limited power. The primary objective of this research is to apply caching strategies that can decrease how often data is sent during the learning process. This involves investigating caching techniques to send only the most crucial model updates, thus cutting down on unnecessary network traffic and boosting the efficiency of the system. The study also looks at how these strategies affect the performance and precision of the learning models, making sure that reducing communication does not harm the quality of the machine learning results. Furthermore, this thesis looks at the wider effects of better communication efficiency in FL for IoT. By lowering the strain on network resources, more devices can take part in the learning process. This could increase the variety and amount of data that helps train the models without risking user privacy. This is especially crucial in areas like healthcare where processing data locally and only sharing vital information can greatly increase user trust and the adoption of the system. Through detailed experiments and assessments, this research seeks to develop a set of guidelines and recommendations for using caching in FL setups. The findings are expected to offer valuable insights into how FL can be made more scalable, efficient, and effective in IoT settings, helping to expand its use in both the commercial sector and academic studies.