Abstract: Secure multiparty computation (MPC) lets parties compute on private inputs without revealing them. For long computations, a fixed set of servers is fragile: servers go offline, and a set known in advance is an easy denial-of-service target. Dynamic-committee models, such as Fluid MPC and YOSO MPC, pass the computation between small, changing committees, but information-theoretic protocols guaranteeing output delivery need communication growing as a high power of the committee size. After introducing the setting, this talk presents three steps towards practicality: letting committees interact for a few rounds before handover, cutting this cost; separating preprocessing and checkpointing from computation, so it resumes from the last valid state after failures; and using homomorphic secret sharing to make communication sublinear in the circuit size without fully homomorphic encryption.

Host: Prof. M. Backendal