GWO-Boosted Multi-Attribute Client Selection for Over-The-Air Federated Learning

Federated Learning (FL) has gained popularity across various industries due to its ability to train machine learning models without explicit sharing of sensitive data. While this paradigm offers significant advantages such as privacy preservation and reduced communication overhead, it also comes with several challenges such as deployment complexity and interoperability issues, particularly in heterogeneous scenarios or resource-constrained environments. Over-the-air (OTA) FL was introduced to address those challenges by sharing model updates without the need for direct device-to-device connections or centralized servers. However, OTA-FL induces some issues related to increased energy consumption, wireless channel variability, and network latency. In this paper, we propose a multi-attribute client selection framework using the Grey Wolf optimizer to limit the number of participants in each round and optimize the OTA-FL process while considering the energy, delay, reliability, and fairness constraints of participating devices. We analyze the performance of our client selection approach in terms of model loss, convergence time, and overall accuracy. Our experimental results show that the proposed multi-attribute client selection can lower energy consumption by up to 43% compared to the random client selection method.