PROJECT TITLE :

Imitation Learning Enabled Task Scheduling for Online Vehicular Edge Computing

ABSTRACT:

The term "vehicular edge computing" (VEC) refers to a potentially useful paradigm that is based on the Internet of vehicles and aims to provide computing resources to end users while also relieving cellular networks of the heavy burden of traffic. In this study, we take into account a VEC network that exhibits dynamic topologies, unstable connections, and unpredictable movement. Offloading computation tasks from vehicles inside the system to available neighboring VEC clusters formed by onboard resources is one way to reduce the overall energy consumption of the system while also meeting the latency requirements of individual tasks. For the purpose of online task scheduling, the majority of recent studies have either developed heuristic algorithms or made use of Machine Learning, such as deep Reinforcement Learning (DRL). However, due to their poor searching efficiency and slow convergence speeds for large-scale networks, these algorithms do not perform well enough to be considered efficient. Instead, we suggest an online task scheduling algorithm that utilizes imitation learning and has near-optimal performance right from the beginning of the process. In particular, a knowledgeable person can obtain the best scheduling policy by solving the formulated optimization problem offline with a few examples. This can be done in a non-real-time setting. We train agent policies for Online Learning by replicating an expert's demonstration while maintaining a level of performance gap that is considered acceptable in theory. The performance results demonstrate that our solution has a significant advantage, with an improvement of more than 50 percent in comparison to the benchmark.