Adaptive joint configuration optimization for collaborative inference in edge-cloud systems

In this study, we propose an adaptive edge-cloud collaborative inference framework that can adaptively configure data and model versions according to task requirements, and decide to transfer them to the cloud server or edge server for inference. Considering the complexity of the joint optimization problem, we decompose the original problem into two low-complexity subproblems. We then propose an adaptive two-stage robust optimization algorithm that can optimize the cost of inference tasks under the accuracy constraint. In the future, we plan to study adaptively edge-cloud collaboration strategies based on feature analysis and content preference awareness.