Federated Few-Shot Zone Learning for Indoor Localization

Indoor localization using fingerprinting has gained attention due to its practicality, but conventional fingerprinting methods require extensively labeled data, making large-scale deployment challenging. Few-shot learning (FSL) enables generalization from limited samples, but its application to indoor localization remains difficult due to data scarcity, distribution shifts, and privacy concerns. This paper introduces Federated Few-Shot Zone Learning (FFZL), the first framework to integrate federated FSL into fingerprinting-based indoor localization. FFZL leverages prototypical networks within a federated setting, allowing decentralized devices to collaboratively classify unseen zones with minimal labeled data while preserving privacy. To further enhance localization accuracy, we introduce FFZL-Map, an improved variant that incorporates spatial regularization into the loss function with the help of the floor map, ensuring learned representations align with the environment's geometric structure. We evaluate FFZL and FFZL-Map on both simulated and real-world datasets, demonstrating the effectiveness of our approach in adapting to novel zones with minimal supervision.