Published Paper Details:

Federated learning allows several devices to train decentralized models while preserving local data, however because of the frequent and significant model changes, it poses significant bandwidth challenges. In real-world applications, scaling these systems necessitates efficient bandwidth management. Numerous tactics are examined in this study, such as compression methods, network innovations, and adaptive protocols. To reduce update sizes without compromising accuracy, we look at methods including model pruning, quantization, and sparse updates. Faster data transfers are made possible by network developments like hierarchical federated learning and asynchronous communication, while adaptive protocols adjust communication based on client accessibility and network conditions. By looking at current advancements and trends, we provide a summary of the literature, identify gaps, and suggest creative ways to increase the efficacy and scalability of FL systems. The objective of this work is to direct further investigations into improving bandwidth management strategies for federated learning.