Towards Vox Populi in Federated Learning: A Fair and Inclusive Client Selection Framework

Federated learning (FL) enables distributed model training without centralizing raw data, but existing client selection methods frequently marginalize resource-limited devices. To address this imbalance, we propose a novel Stackelberg-game-based framework that shifts decision-making from a server-centric model to a client-to-client paradigm. Resource-rich leaders form coalitions with resource-limited followers, allowing every client, regardless of capacity, to contribute to and benefit from the global model. We further introduce a comprehensive data scoring mechanism that evaluates the quality and diversity of each client's dataset, ensuring that coalition formation is both fair and inclusive. Experiments on diverse benchmarks demonstrate that our approach outperforms the state-of-the-art, as well as different baselines on multiple metrics while promoting fairness. Our proposed strategy effectively tackles heterogeneous data distributions and uneven resource availability, offering a scalable, equitable solution for real-world FL scenarios.