Multi-Party Computation

Multi-Party Computation(MPC) enables multiple parties to collaborate on computations without disclosing their data to each other. Based on MPC protocols, multiple servers can jointly train a model by secret sharing. However, this approach incurs high computational and communication overhead, especially for large models, which require significantly more computation and communication resources. Moreover, MPC protocols necessitate simultaneous coordination of all parties, which may contradict the decentralized nature of environments where parties are unreliable.