Using Qbeast Spark to make the Machine Learning models more efficient.
In January, 2018 the I-BiDaaS project* was successfully launched with 13 participating organizations from 8 different countries. The Barcelona Supercomputing Center (BSC) was a member of the I-BiDaaS consortium, and Qbeast’s solution took part in the I-BiDaaS platform as BSC’s contribution. Telefonica participated as the organization leading the telecommunications use case, with the purpose being the use of big data to predict congestion points in the cellular network and employ bots in call centers that are more anthropomorphic and realistic.
After a successful cooperation between Telefonica and Qbeast Ioannis Arapakis, Research Scientist at Telefonica with Adrià Correas, Junior Engineer at Qbeast are working together on a follow-up project which addresses the high-value problem of cellular network workload forecasting using Recurrent Graph Neural Networks.
Rapid urbanization brings with it an influx of people into cities, posing enormous mobility and sustainability issues. With the introduction of IoT-based technologies to the market and the broad adoption of mobile devices by end-users, the growing need for telecommunication services is being exaggerated even more.
Currently, billions of mobile users access the Internet while on the move. Because it is the only global-scale infrastructure with ubiquitous mobility support, the cellular network is critical for their Internet access. A high-value service is at the heart of its mobility management: performance predictions of workloads predicted to be generated in the near future.
Graph Neural networks are a hot topic right now, as they allow us to train Machine Learning models with information about the whole structure of a graph, not just node features. What Ioannis and Adrià are trying to do is improve the previous approaches in order to solve this problem using this technique. The Qbeast Spark extension will push this approach forward as it can help make this computationally expensive task more efficient.
What is proposed is the creation of a centralized network solution to handle the aforementioned challenge in this project. Typically, there are hundreds of thousands of sectors providing tens of measurements at a constant (almost instantaneous) rate, making accurate and timely forecasting of congestion points particularly challenging, which makes it a valuable service to mobile network operators. Ioannis and Adrià will identify and quantify the most prominent patterns in a cellular network and research a solution, which can be executed in a cloud environment that can handle resource-hungry AI algorithms.
*The I-BiDaaS project has received funding from the European Union’s Horizon 2020 Research and Innovation program under grant agreement No 780787.