Mobile networks are experiencing explosive growth in complexity, due to densification (growing number of cells) as well as New Radio (more complex cells). At Swisscom, we expect a 2.5x increase in complexity of our optimisation effort, which can only be addressed effectively and efficiently through leveraging ML models. Hence, we developed a data product called AutoTune (no, not the music one!). It uses configuration and performance data to predict future cell performance and automatically improve it. This talk explores its architecture, the models considered and lessons learned during its development. We show that ML can have a real impact on network performance, as well as improve the efficiency of domain expert teams.

Download the slides for this talk.Download ( PDF, 2842.88 MB)