Federated Learning: A Literature Review on Decentralized Machine Learning Paradigm

Document Type : Original Article

Authors

1 Information System Department ,Faculty of Computer and Artificial Intelligence , Helwan University , Cairo, Egypt

2 Department of Information Systems, Faculty of Computers and Artificial Intelligence, Helwan University, P.O.Box 11795, Egypt.

3 Information System, Faculty of Computer and Artificial Intelligence, Helwan University, Cairo, Egypt

Abstract

Federated Learning (FL) refers to a groundbreaking paradigm for distributed machine learning (ML), ensuring model training without compromising the privacy of local data. Despite its promise, FL suffers from some challenges, involving concerns over direct data leakage, the potential of compromising the model architecture privacy, and the overheads associated with connection and communication. This paper shows an in-depth study of FL, and its categorization according to the data partitioning formats such as horizontal FL, vertical FL, and federated transfer learning. A thorough examination of FL models is given, highlighting the need to incorporate strong privacy and security protections inside FL frameworks and illuminating the inherent difficulties these models present. The paper also examines previous research on FL, on how integrating security techniques to improve FL systems' general effectiveness. By consolidating current knowledge, the paper provides a roadmap for future directions, highlighting the possible solutions in mitigating challenges and advancing privacy-preserving federated learning.

Keywords

Main Subjects

Files

Share

How to cite

Statistics