Efficient Training Data Gathering for Federated Learning in Multi-Hop Wireless Network

Machine learning performance strongly relies on the quality and quantity of training data sourced from various origins to create reliable and efficient models. In distributed learning paradigms such as Federated Learning (FL), wireless multi-hop networks (WMNs) have significant potential to play a crucial role in effectively gathering data for multiple servers. However, these networks face substantial obstacles, including interference, fluctuating bandwidth, and ineffective routing, especially when numerous source nodes transmit data to several servers until all storage capacities are filled. In wireless mesh networks, our goal is to identify methods for constructing well-formed dynamic multicast groups that ensure essential training data reaches servers effectively, thereby minimizing the estimated time required for all storage units to reach capacity. Experimental results demonstrate improved data collection efficiency and reduced network congestion compared to traditional routing approaches in NS3 simulator.