A multi-party logistic regression using the Adam method

Secure Multi-Party Computation (MPC) allows multi-party to jointly compute a function while ensuring data security and computation correctness. Nowadays, due to the lack of trust among data holders, various inconveniences arise in data sharing. Secure multi-party computation can partially address such issues by securely enabling computations involving multiple parties’ data while protecting the privacy of each party’s data. In this paper, a secure multi-party Adam method is proposed based on secure multi-party computation, and the non-linear operations in the original scheme are approximated. By optimizing the computation of the power function in the secure multiparty computation protocol, the protocol can compute higher-order Taylor expansions with lower communication complexity and ensuring security. Through the optimization of the secure multi-party computation protocol itself, the proposed multi-party Adam method achieves lower communication overhead and faster convergence speed compared to traditional multi-party gradient descent methods within the same number of iterations.