Accuracy-Based Task Offloading and Resource Allocation for Edge Intelligence in IoT

Machine learning (ML) tasks in Internet of Things (IoT) are sensitive to task inference accuracy. In this letter, an ML task offloading scheme is proposed to minimize the total delay of task processing in an edge-intelligence-enabled IoT scenario, while guaranteeing the accuracy requirements of tasks, and taking into account the multiple attributes of tasks, task inference accuracy, and impact of error inference on task processing delay. The problem of wireless channel allocation, and computing resource allocation is modeled along with the task offloading. Considering the high complexity of the optimization problem, we design an algorithm which decomposes the problem into a computing resource allocation sub-problem and a task offloading and channel allocation sub-problem, and then solves them separately. In extensive simulations, the superiority of our scheme is demonstrated in comparisons with 4 other schemes.