Special session - Emerging research and applications at the frontier

As first proposed by a multidisciplinary academic group two decades ago, attenuation measurements from microwave backhaul links can be used to derive rainfall information. Since then, various research groups have employed AI-based methods to extract spatial rainfall observations from wireless communication networks. These approaches have strong potential to complement conventional monitoring systems and help fill observational gaps, particularly in data-scarce regions such as low- and middle-income countries. Although mobile network operators routinely collect these measurements for network operation and maintenance, their use in meteorology and hydrology has so far remained limited to a small number of research projects, mainly due to data-sharing challenges.

In this talk, I will introduce the Global Microwave Link Data Collection Initiative (GMDI), which focuses on establishing a sustainable global data flow framework for the benefit of all stakeholders. In particular, the EU-supported SetGMDI project is developing pilot activities and engaging with the key stakeholders on whom GMDI depends: mobile network operators, national meteorological and hydrological services, and other rainfall end-users across Europe and Africa. Real data is already being collected from European networks, while African pilot deployments are in their final stages of preparation, including the completion of legal and data-sharing agreements.

The talk introduces the concept of the deployment gap: the structural disconnect between how AI systems are evaluated before deployment and how they behave in the real world. Despite impressive benchmark scores, AI systems routinely fail to deliver reliable performance when moved into operational settings. The deployment gap can be understood along four interacting dimensions: the data gap, the evaluation gap, the system gap, and the society gap. Attendees will understand that reliable claims about AI safety and effectiveness must be grounded in actual use conditions, not approximations.

In this presentation, we introduce an architectural framework that integrates artificial intelligence with edge and fog computing to enable autonomous operation in next-generation networks, including IMT-2030. Drawing from the Hierarchical Autonomous Edge Architecture, Edge Sandbox addresses a core challenge: delivering sophisticated self-management capabilities across highly heterogeneous fog/edge nodes, including those with severe resource constraints. By enabling local, sandboxed evolution and hierarchical orchestration across cloud, fog, and edge layers, this architecture substantially reduces the load on the core network while improving the efficiency of distributed learning. Processing data locally minimizes latency, conserves bandwidth, and reduces inter-network communication volume—essential for scalable and responsive autonomous operations. The Edge Sandbox supports advanced applications such as telepresence, holography-type communications, autonomous vehicles, industrial IoT, and remote medical monitoring, enabling immediate and reliable responses to real-world events in autonomous network environments.

Celina Lee, Zindi co-founder and CEO
Talk: Zindi: Building AI Research and Innovation Capacity Across Africa
Abstract