PartialLoading: User Scheduling and Bandwidth Allocation for Parameter-sharing Edge Inference