Research on Efficiency and Security Balance of Federated Learning Algorithm for Multi-node Privacy Computing

In order to solve the key problem of improving algorithm efficiency while ensuring data security, this paper introduces an innovative method, the cryptographic-Homomorphic Asynchronous (EHA) method, which combines homomorphic encryption (HE) with asynchronous federation learning (AFL). EHA method uses HE to realize the encrypted transmission and calculation of data, ensuring the confidentiality of data in the whole process and preventing information leakage. At the same time, AFL is used for asynchronous model update, allowing nodes to update model parameters at different time points according to their own conditions, effectively reducing the delay caused by synchronous waiting, and significantly improving the efficiency and response ability of federation learning. Finally, by comparing with the traditional federated learning algorithm, experiments and analysis of experimental results show that compared with the traditional methods, EHA has significant advantages in communication efficiency, confusion matrix optimization and system robustness enhancement. Especially in the multi-node environment, EHA method not only effectively balances the relationship between efficiency and security, but also significantly reduces the communication delay and computing cost, and significantly improves the overall performance of the system. These results fully verify the feasibility and effectiveness of EHA architecture in the federated learning domain of multi-node privacy protection.