Mapping the unseen: Robust IP geolocation through the lens of uncertainty quantification

Accurate IP geolocation is critical for applications such as network security, content delivery, and fraud detection, yet existing methods face significant challenges in dynamic environments with fluctuating network conditions. In this work, we present EBGeo (Energy-Based IP Geolocation), a novel framework that combines graph convolutional networks (GCNs) and energy function optimization with Monte Carlo sampling to address these challenges. The proposed framework introduces three key innovations: (1) GCNs, which model the spatial and topological relationships between IPs and are well suited to the IP geolocation task by capturing complex dependencies in network structures; (2) energy-based optimization, which leverages energy function optimization with Monte Carlo sampling to simulate dynamic network conditions during training, thereby enhancing the model's accuracy and robustness; and (3) gradient ascent for inference, which improves the model's adaptability under fluctuating network conditions. Uncertainty quantification (UQ) is used to evaluate how well the model adapts to network changes. Lower UQ values indicate that the model is less sensitive to variations in network conditions. UQ further enables a deeper understanding of the model's adaptability to changing network conditions, making EBGeo a powerful tool for addressing network challenges in real-world applications.