Federated Distillation Based Indoor Localization for IoT Networks

The federated distillation (FD) paradigm has been recently proposed as a promising alternative to federated learning (FL), especially in wireless sensor networks with limited communication resources. However, all state-of-the-art FD algorithms are designed for classification tasks only, and less attention has been given to regression tasks. To take advantage of this promising paradigm for localization which is inherently a regression problem, we introduce in this work an FD framework that is tailored to address regression learning problems. More specifically, we propose an FD-based indoor localization system that shows a good tradeoff in communication load versus accuracy compared to FL-based indoor localization. With our proposed framework, we reduce the number of transmitted bits by up to 98%. We analyze the energy efficiency regarding the number of communication rounds for both FD and FL systems to achieve the same localization accuracy. The results revealed that FD comes with greater energy efficiency by reasonably saving transmission energy at the expense of computation energy. This notably represents a significant advantage in the context of wireless sensor networks, particularly given the prevalence of battery-powered Internet of Things (IoT) devices operating with constrained bandwidth. Moreover, we show that the proposed framework is much more scalable than FL and, thus more likely to cope with the expansion of wireless networks.