DNN-Based Task Partitioning and Offloading with Reliability Guarantees in Multi-UAV-Assisted MEC System

In recent years, artificial intelligence (AI) applications based on Deep Neural Networks (DNN) have gained widespread attention in mobile edge computing (MEC). However, it is still challenging to ensure reliable transmission and meet delay-sensitive requirements in dynamic networks. In this paper, a multi-UAV-assisted MEC architecture is designed, where multi-ple UAV s are endowed with computing offloading services for the mobile devices (MDs). To guarantee reliable communication in highly dynamic scenario, we formulate the robust transmission problem, and leverage the conditional value-at-risk approach to analyze the successful transmission probability in the worse-case channel condition. On this basis, to accelerate the inference of DNN, we formulate the Collaborative DNN Partitioning and Offloading (CDPO) problem by presenting a collaborative DNN partitioning and offloading scenario, which is NP-hard. Further-more, we introduce a highly effective Collaborative Partition and Offloading Optimization (CPOO) algorithm based on the Bat algorithm to solve the CDPO problem. In addition, considering both reliability-guaranteed and delay-sensitive requirements, we propose the master problem as a non-convex optimization prob-lem and the LaCPOO algorithm, which combines the sub gradient algorithm with the CPOO algorithm. Simulation results demon-strate that compared to other algorithms, our proposed approach can accelerate DNN inference while ensuring reliability.