Model Self-Supervised Federated Learning for Heterogeneous Fingerprint Localization

Fingerprint localization is emerging as one of the most important indoor positioning methods, especially with the development of deep learning (DL). To avoid potential privacy leakage of clients from the large-scale fingerprinting samples for DL training, federated learning (FL) has been introduced to implement distributed learning without exchanging raw local data. However, FL-based localization suffers from severe accuracy degradation caused by heterogeneity among distributed data. To tackle this problem, this letter builds a model self-supervised federated learning (MSS-FL) framework for fingerprint localization. In MSS-FL, we propose to conduct self-supervised learning at the model level, and incorporate a model self-supervised loss into the local training objective to relieve the bias between global model and local model. Experiments on a real-world dataset are conducted to validate the improved localization accuracy with faster convergence of MSS-FL, compared with other state-of-the-art FL methods.