Edge-enabled Consumer Digital Twins in Industrial Metaverse
Yue Han, Wei Yang Bryan Lim, Dusit Tao Niyato, Cyril Leung, Chunyan Miao · 2023
Industrial Metaverse has recently attracted considerable attention from both academia and industry due to its real-world mirroring, decentralization, and interoperability for enabling next-generation Internet services such as virtual consumer banking and retail. Consumer digital twins, i.e., the digital replica of a consumer, can be important in the industrial Metaverse, since multiple sales strategies, consumer engagement interventions, and advertising campaigns can be tested on consumer DT before an actual intervention, significantly increasing business efficiency and profits. A consumer DT may involve multi-dimensional dynamic real-time data (e.g., location and behavioral data) which requires assistance via the IoT devices (such as wearables and mobile phones). However, some information regarding consumers is confidential and private. In this paper, we propose a edge-based privacy-preserving training framework for developing consumer DTs. In particular, a global pre-trained consumer DT is placed at some edge clouds which linked to multiple base stations (BS). Wireless devices that provide confidential consumer data can download the model, finetune it with local data, and transmit updated model parameters to the edge server via a selected BS. The parameters of the consumer DT are to be aggregated at the edge cloud hosting the global consumer DT and the timing of update is affected by the network congestion. We adopt evolutionary game to model the opportunistic BS selection behavior of the IoT devices. Extensive numerical simulations and theoretical analysis show that in the long-term a equilibrium point can be achieved.