Towards Cognitive Autonomous Networks

6Machine Learning: Tools for End‐to‐End Cognition

Stephen Mwanje¹, Marton Kajo², and Benedek Schultz³

¹Nokia Bell Labs, Munich, Germany

²Technical University of Munich, Munich, Germany

³Nokia Bell Labs, Budapest, Hungary

Network management (NM) is responsible for processing network and service‐ related information, to draw insights about the network and its related services and to derive network and service reconfigurations in order to optimize resource usage while minimizing cost. Traditionally, NM tasks were undertaken by human experts who interpreted network events and decided the best course of action. As networks became increasingly dense and complex, supporting a higher number of experts needed to manage the networks became rather expensive. The Self‐Organizing Networks paradigm solved the challenge by automating the expert‐like decision making through classic AI techniques, such as expert systems or rule‐based closed control loops. However, as discussed in Chapter 5, classic AI techniques have major limitations – mainly, the inability to achieve end‐to‐end cognition owing to the need to explicitly describe the models of the world. Thereby, Machine Learning (ML) is the solution.

End‐to‐end cognition is expected to improve NM automation by reducing the need for explicit models underlying the NM decisions. For example, it should no longer be necessary for the operator to describe the full set of possible faults, the sequences in which they occur and the related solutions. ...