Accumulative Fidelity Maximization of Inference Services in DT-Assisted Edge Computing

Digital twin (DT) technology illuminates the smooth integration of cyber and physical worlds in alignment with the Industry 4.0 initiative. Through synchronizations with physical objects, DTs of objects can reflect the states of the objects with high fidelity. Machine learning-based service models through continual training by the DT update data can provide users with accurate inference services. Orthogonal to the DT technology, mobile edge computing (MEC) is a promising computing paradigm that shifts computing power to the edge network, which is particularly appropriate for delay-sensitive intelligent services. In this paper, we study the fidelity enhancement of service models in a DT-assisted edge computing environment empowered by 6G communication, through continuously training service models, using DT update data obtained from their mobile $\mathbf{IoT}$ devices (objects) in a real-time manner. To this end, we first formulate an accumulative fidelity maximization problem that jointly considers the placement of DTs and models, with the aim to maximize the accumulative fidelity gain of all models while minimizing the total cost of resource consumed due to DT updating and service model training. We then develop an efficient algorithm for a sub-optimization problem - the placement problem of DTs and models, under the assumption that the mobility profile of each mobile object is given. When both DTs and models have already been placed, we devise an algorithm for the accumulative fidelity maximization problem that schedules each object to choose an access point (AP) to upload its update data at each time slot for a given time horizon to maximize the accumulative fidelity of all service models while minimizing the total cost of various resources consumed. We finally evaluate the performance of the proposed algorithms through simulations. Simulation results indicate that the proposed algorithms are promising, and outperform their baselines nearly by 20%.