Martin: Mobility-Aware Reputation Mechanism for Federated Learning

Abstract

The rapid development of the Internet of Things (IoT) has resulted in an increasing volume of data, much of which contains sensitive private information. Federated Learning (FL) allows clients to train models without sharing raw data, showing significant potential for privacy protection. However, device mobility presents challenges for applying FL in IoT, including high communication costs, data quality fluctuations and intermittent device participation. We propose a mobility-aware reputation mechanism for federated learning (named Martin) to address these issues. First, we designed a quality evaluation method based on sparse gradient similarity, which evaluates the data quality of devices while reducing communication costs. Additionally, we designed a dynamic reputation mechanism to quantify the quality evaluation records of devices, enabling the correct selection of intermittently participating devices from a dynamically changing set of devices, thereby improving model training quality under limited communication bandwidth. Experimental results validate its effectiveness.