Model-Preserving MPC with Low Communication Based on Multivariate Nonlinear Problems

Polynomial models are essential for capturing non-linear quantitative relationships across various fields. As demands grow for robust protection of both private data and proprietary models, most secure multi-party computation (MPC) protocols- which mainly focus on data privacy-prove inadequate for model protection. We propose a model-preserving MPC protocol that guarantees complete privacy for polynomial models while significantly reducing communication overhead. By enabling secure computations with low communication, our approach minimizes both the communication overhead and the risk of collusion attacks. Notably, our protocol requires the parties to perform two rounds of communication.