The Evolution of Machine Learning: From Centralized to Distributed
Jayakrushna Sahoo, Akarsh K. Nair, Richa Sharma · Apple Academic Press eBooks · 2024
Motivated by the increasing concerns over privacy and the gradual growth of newer technologies such as IoT, the past few years have witnessed a shift in the perception of machine learning from centralized to distributed. An emerging distributed learning paradigm, known as federated learning (FL), is gaining popularity due to its features such as increased privacy preservation, data locality, distributed framework, and remote training. The FL network architecture is based on a client-server system, where data is fully present at the client end and the system training takes place there as well. This eliminates privacy and security concerns while ensuring efficient communication. This article initially presents various terminologies and classifications associated with traditional machine learning, and then discusses the need for a paradigm shift. We also provide an overview of the distributed learning literature, followed by an introduction to federated learning. The article further explores common applications, various network architectures, and detailed use cases. Finally, the article concludes with a discussion of some of the open challenges and future research directions in the field of FL.