SCOPE-FL Enhances Federated Learning with Strategy-Proof, Pareto-Efficient Client Selection.

Hierarchical Federated Learning (HFL) offers a scalable approach for collaborative model training across numerous distributed devices while maintaining data privacy. However, current HFL client selection mechanisms often fall short due to strategic inefficiencies. These systems frequently prioritize stability over Pareto efficiency, leading to suboptimal resource allocation. Additionally, without strategy-proofness, participants are incentivized to misrepresent their true preferences, which degrades the overall system welfare. To overcome these limitations, researchers propose SCOPE-FL (Strategy-proof Chain-based Optimal Pareto Efficient Federated Learning). This new synchronous HFL framework redefines client selection as a two-sided school choice problem, which is then solved using the Top Trading Cycle (TTC) algorithm. This approach simultaneously guarantees both Pareto efficiency and strategy-proofness. For distributing rewards, SCOPE-FL employs a scalable approximation of the Shapley value, based on One-Round Reconstruction (OR), ensuring that compensation is directly proportional to each client's contribution. The entire mechanism is executed via blockchain smart contracts, providing a secure and tamper-proof environment essential for upholding the strategy-proofness guarantees. Extensive evaluations on datasets like MNIST, Fashion-MNIST, and CIFAR-10 demonstrate that SCOPE-FL significantly outperforms existing state-of-the-art methods in terms of model accuracy, convergence rate, and reward efficiency, while maintaining competitive communication latency and lower blockchain overhead.