يادگيري مشاركتي بر پايه زنجيره قالب عمومي از طريق تراكنش هاي خارج از زنجيره

Today, artificial intelligence and consequently machine learning have found extensive applications in industries and daily life. However, traditional machine learning methods face challenges such as data centralization, privacy violations, and inefficiency at large scales. Collaborative learning, as an innovative approach, aims to overcome these challenges by distributing the learning process and achieving a high-quality final model through data distributed among various devices. In this research, a protocol for collaborative learning based on general blockchain templates is presented, offering multiple advantages. Firstly, by utilizing general blockchain templates, we can access a large population of users with diverse data distribution, ensuring transparency and eva‎luation of the learning process. Secondly, off-chain transactions enable efficient exchange of data and models without the need to record transactions on the blockchain. Thirdly, using incentive mechanisms, users are encouraged to actively participate in the learning process. On the other hand, due to the transparency and distributed eva‎luation by users, it is possible to penalize malicious users through a smart contract. This protocol, focusing on privacy preservation and efficiency, effectively addresses existing challenges in collaborative learning such as fairness mechanisms, reward and penalty mechanisms, and decentralized eva‎luation. It provides a suitable foundation for the development and widespread application of this method in various fields. The design of this protocol is carried out in seven phases. Finally, the protocol is simulated using the MNIST dataset and the Python language, and based on this simulation, conclusions have been drawn.

5