Efficient Federated Digital Twin Synchronization in Edge-Cloud Collaborative System

Ji-Wan Kim, Hong Je-Gal, Hyun-Suk Lee

Abstract: In this paper, we study a federated digital twin synchronization in an edge-cloud collaborative system with distributed DNN (DDNN) and early exit structures. The federated digital twin is composed of multiple digital twins whose synchronization is performed by using deep neural network (DNN) models. In the system, each DNN model for synchronization is partitioned into two sets of layers that are executed at the edge and the cloud respectively. This enables the system to perform efficient synchronization by utilizing the computing resources of the system effectively. To fully exploit this advantage, we formulate a partition point optimization problem to minimize the expected total time consumption for the federated digital twin synchronization considering the early exit structure of the DNN models. Then, we propose an exhaustive partitioning algorithm that identifies the optimal partition points of the DNN models for synchronization. We also propose a heuristic algorithm that finds the suboptimal partition points, but has much lower computational complexity compared with the optimal algorithm. Through the simulation, we demonstrate that our proposed algorithms can effectively reduce the total time consumption for federated digital twin synchronization.