Distribution-Aware Mobility-Assisted Decentralized Federated Learning

Decentralized Federated Learning (DFL) eliminates reliance on a central server by enabling peer-to-peer model aggregation, but suffers from limited connectivity and non-IID data. In many applications, some clients are mobile, yet the role of mobility in DFL remains underexplored. This work shows that introducing even a small fraction of mobile clients, moving randomly, significantly improves information flow and accuracy. To further enhance performance, we propose structured distribution-aware mobility strategies, where mobile clients navigate the network using knowledge of data distributions and static client locations. These strategies mitigate the impact of data heterogeneity and accelerate convergence. Extensive experiments validate the effectiveness of induced mobility in DFL and demonstrate the superiority of our proposed mobility patterns over random movement.