Communications networks are an important lifeline of our society. In order to keep up with the enormous technological progress, they will be operated with the support of Artificial Intelligence (AI) and Machine Learning (ML) in the future. “Especially with regard to real-time networks, where transmission guarantees must be given and adhered to, good and rapid prediction of delays in the backbone network is important,” says Peter Dorfinger, Head of the Intelligent Connectivity department at Salzburg Research. “With the help of reliable predictions, network operators can quickly reconfigure their network or test various network configurations.”
Numerous organizations worldwide are researching how Artificial Intelligence/Machine Learning (AI/ML) can best be used in future networks, including 5G networks. In an international challenge, 911 teams from 62 countries competed with their know-how in 23 individual challenges. The best teams were invited to the final in late December 2020. Salzburg Research achieved an excellent second place across all 23 challenges.
Gold went to China Mobile. Among the other participants were TU Berlin, Imperial College London, Eurecom, Chinese Academy of Sciences, Inria, Fraunhofer HHI, NEC, IMDEA and many more.
Efficient AI solution
The research team primarily achieved second place through the efficiency of the developed AI/ML solution: “While the solutions of other teams had to learn for a whole week, the AI from Salzburg Research was only trained for 48 hours and still predicted the delay excellently,” says Dorfinger. Another decisive factor for success is the interdisciplinarity in the research team, consisting of mathematicians and computer scientists. “Our knowledge of networks enabled us to keep the complexity of the system low. The expertise from mathematics provided additional performance,” Dorfinger continues.
The research and development work of the Salzburg Research researchers Martin Happ, Christian Maier, Jia Lei Du and Matthias Herlich was carried out within the framework of the IDA Lab Salzburg.