Machine Learning for Network Automation: Overview, Architecture, and Applications [Invited Tutorial]
This tutorial is an excellent introduction to the major concepts and applications of machine learning (ML) in optical networks. It presents an overview of the three main ML approaches (supervised learning, unsupervised learning, and reinforcement learning), including the salient characteristics, algorithms, and applications of each. In addition to introducing the relevant terminology, this provides a foundation for understanding how ML works, its strengths, as well as its limitations. As data is the driver of ML, the tutorial covers the inherent challenges posed by data management, including monitoring, storage, and representation. One of the distinguishing contributions of the tutorial is a detailed description of the network management architecture, to illustrate how ML can be integrated into existing network software stacks. The tutorial also includes an in-depth look at the application of ML in four important use cases: predictive maintenance, virtual network topology management, physical layer capacity optimization, and optical spectrum analysis. Overall, the tutorial serves as an excellent means to "get up to speed" on a technique that is expected to be transformative in optimizing and operating optical networks, and networks in general.