Federated Learning for Privacy-Preserving Optimization in Multi-Domain Optical Networks

Abstract

Modern optical networks have been seriously challenged by new infrastructures and data intensive applications that have increased to an alarming rate. Intelligent, scalable, privacy-preserving optimization mechanisms are needed in multi-domain optical networks (MD-ONs), which entail multiple administrative and technological domains. Federated Learning (FL) is a promising novel decentralized machine learning model capable of addressing all these challenges by offering the possibility to jointly train models without sharing raw data. The work in this research paper explains how federated learning could be integrated to the multi-domain optical network to enable privacy-preserving optimization. It talks about architectural buildings, optimization techniques, communication boundaries, and practical applications, problems and research gaps. The paper sheds light on the ways in which FL can improve the efficiency, scalability and privacy of data on networks, and this makes it one of the enabling factors in the next generation of intelligent optical networks.