Game-Aided Blockchain Twin for Incentive and Relay-Free Model Sharing in Heterogeneous Chain-Driven Swarm Learning

Swarm learning (SL) is a novel decentralized machine learning paradigm that provides a privacy-preserving approach based on permissioned blockchain without the need for a centralized coordinator. However, the various architectures and design characteristics of blockchains make it difficult to employ applications on heterogeneous blockchains, which limits the scalability, efficiency, and interoperability of blockchains ecology and restricts the application of SL. To solve this problem, first, we propose a Blockchain Twin mechanism consisting of multichains to enable model sharing between heterogeneous blockchains without single central relay-chain. Next, to encourage roles in Blockchain Twin to actively and honestly participate in consensus phase, we design a multileader multifollower Stackelberg game-based incentive mechanism. Additionally, we prove that a unique Stackelberg equilibrium exists in the game and propose an alternating direction method of multipliers (ADMM)-based algorithm to obtain the optimal solution. Finally, we evaluate the performance of twin-chain interactions regarding average delay and throughput. We also conduct numerical simulation on the proposed incentive mechanism, and the results show that our mechanism can jointly maximize the reward of every participant roles in Blockchain Twin.