Deep-learning Based Auction Resource Allocation in Coded Computation Offloading for Internet-of-Things

Abstract

Edge computing has been a promising approach that allows resource-constrained Internet-of-Things (IoT) devices to offload their computation-intensive tasks to the edge of the networks. Instead of offloading the entire computation task to a single edge device, the IoT devices can leverage on the resources of multiple edge devices. In order to improve the performance of the distributed computation tasks, coded distributed computing is proposed to mitigate the straggler effects. However, the edge devices may not have sufficient bandwidth to transmit the computed results to the IoT devices reliably. Hence, we present a deep-learning based auction for the edge devices to buy and utilize bandwidth from the service providers. In particular, the edge devices compete for the bandwidth of the service providers and the service providers allocate bandwidth to the edge devices that value it most while maximizing their revenue. The simulation results show that the deep-learning based auction is optimal as it maximizes the revenue of the service providers while satisfying both properties of individual rationality and incentive compatibility.