AI Task Allocation and Inference Offloading in Edge–Cloud Systems

AI applications increasingly rely on edge–cloud infrastructures to meet strict latency and freshness requirements. This thesis presents a survey of distributed AI task allocation and inference offloading strategies in edge–cloud networks under timeliness constraints. The analysis is organized by categorizing existing works according to server selection, task allocation, inference awareness, freshness metrics, and centralized versus distributed decision models. The survey highlights common trade-offs between latency, freshness, and system efficiency, and shows that distributed approaches often achieve near-optimal performance under realistic conditions.