The Euclid team took part in the “Machine Learning in 5G Challenge” competition held under the auspices of the International Telecommunication Union (ITU) by developing their own solution and presenting the work entitled “Federated Learning for 5G Base Station Traffic Forecasting”. In particular, the team competed and won 1st place in the challenge concerning the optimization of Federated Learning methods for 5G data prediction. At the same time, in the overall competition (grand challenge) among all the winners of the individual challenges, was ranked 4th.

Problem Description

The goal of the Federated Traffic Prediction for 5G and Beyond Problem Statement was to investigate the usage of Federated Learning tools to predict the traffic in cellular networks collaboratively. To that purpose, we were provided with an unlabeled dataset that contains data from unknown LTE users of commercial operators at three different specific locations in Barcelona, Spain. The main objective of the challenge was to accurately predict the downlink and uplink stream size, using past observations from the given dataset without revealing information between base stations.

Participation

Our team presented a benchmark for Federated Learning for 5G Base Station Traffic Forecasting, where various optimizations to existing methods were introduced. The official Github repository (incl. Presentation Slides) of our solution can be found here.

Watch the Machine Learning in 5G Challenge Grand Finale