The Networks Data Science (NDS) group at IMDEA Networks Institute, led by Dr. Marco Fiore, carries out research at the interface of mobile networking and data science, by applying and tailoring tools from artificial intelligence, machine learning, statistical analysis and data mining to the metadata that flows through modern mobile network architectures. We address research problems on the characterization, modeling and forecasting of the complex dynamics of mobile data traffic, and use the derived insights to improve the design and operation of mobile network architectures. We also leverage the rich information available in mobile network metadata to solve challenging problems in social sciences with a computational, data-driven approach. Our research generally builds on large-scale measurement data collected in the operational systems of major mobile network providers, and is carried out in cooperation with leading academic and industrial partners.

Mobile traffic analysis and generation
We study how mobile services are consumed at large (metropolitan and national) scales, by exploring measurement data recorded in production networks. We aim at unveiling the specific dynamics of traffic demands generated by a variety of apps in space and time, including the complex commonalities and diversities in the way mobile services are used in different times or geographical areas, or the way mobile traffic consumption is affected by special events or circumstances such as the COVID-19 pandemic. We also develop models for the generation of synthetic datasets of spatiotemporal mobile network traffic that can support data-driven research.

Network intelligence
We design data-driven algorithms for the effective, anticipatory and automated orchestration of network resources. To this end, we apply artificial intelligence, machine learning, statistical methods and optimization tools to traffic data collected in large-scale operational mobile networks. We perform assessments of the limits of current and future technologies, such as resource utilization efficiency in network slicing, when confronted with real-world traffic loads. We also develop original solutions for network management hinging on deep learning models, such as predictors of the capacity to be allocated to network slices to minimize operation costs, or lightweight approaches to service classification from traffic aggregates. Our network intelligence solutions are based on custom model designs that contribute to advancing the sate of the art in machine learning for practical engineering tasks.

Remote sensing with mobile network metadata
We take advantage of the rich information embedded in data generated within mobile network architectures to solve problems in social sciences. The nearly ubiquitous coverage and high level of detail of individual and aggregate metadata available to operators allow developing data-driven models that largely improve the current state-of-the-art in disciplines like geography, demographics or economics. As representative examples, we employ mobile network metadata to produce cartographies of mixed land use in urban areas, estimate the dynamics of population density in cities at order-of-minute granularity, or predict the socioeconomic status of urban neighborhoods.