With the advancement of Internet-of-Things (IoT), enormous IoT data are generated at the network edge, incurring an urgent need to push the frontiers of artificial intelligence (AI) to network edge so as to fully unleash the potential of the IoT big data. To match this trend, edge intelligence—an emerging paradigm that hosts AI applications at the network edge—is being recognized as a promising solution. While pilot efforts on edge intelligence have mostly focused on facilitating efficient model inference at the network edge, the training of edge intelligence model has been greatly overlooked. To bridge this gap, in this paper, we investigate how to coordinate the edge and the cloud to train edge intelligence model, with the goal of simultaneously optimizing the resource cost and preserving data privacy in an on-demand manner. Leveraging Lyapunov optimization theory, we design and analyze a cost-efficient optimization framework to make online decisions on training data scheduling to balance the tradeoff between cost efficiency and privacy preservation. With rigorous theoretical analysis, we verify the efficacy of the presented framework.
展开▼
