Energy consumption and efficiency in federated learning (FL) for IoT
Kuldeep Singh Kaswan, Jagjit Singh Dhatterwal, Kiran Malik, K. Babu · 2024
Federated learning (FL) can be therefore described as a revolutionary technique within machine learning deemed specific for the Internet of Things (IoT) context due to its focus on privacy and communication overhead. FL differs from traditional machine learning as it does not centralize data, instead, the training process is performed by multiple IoT devices while simultaneously keeping data local and the global model is learned collaboratively. This method forms a way of solving some problems like bandwidth limitation, energy utilization, delay, expandability, and heterogeneous data in IoT systems. Communication paradigms are critical and sensitive in FL for IoT and therefore require effective management. Loosely coupled IoT networks have been defined to have particular requirements for bandwidth management, energy conservation, and real-time data processing that have motivated better techniques. Issues such as model compression (pruning, quantization), gradient compression, and adaptive communication methods play a central role in improving communication rates. Life management strategies such as energy-conscious communications procedures and optimum distribution of units also contribute to the endeavor.