A unified deep learning framework for energy-efficient offloading and resource allocation in multi-access edge computing

The problem of computational offloading in mobile edge computing is of great importance due to the limited resources of user equipment (UEs). In this paper, we propose parallel execution of multiple tasks on multiple available mobile edge servers (MESs). We formulate a comprehensive cost function which considers time delay and energy consumption of UEs in deciding task allocations to multiple MESs. The minimization problems of UE’s energy consumption and service delay due to computational offloading are solved through the proposed deep learning-based parallel tasks offloading scheme (DPTOS). We formulate the problem of computational offloading and resource allocation in a multi-user multi-server scenario into a simple framework where different objects are placed in different boxes to generate an accurate training dataset. Our cost function generates training datasets having minimum energy consumption and service delay. Performance evaluation in comparison with eminent computational offloading schemes shows that DPTOS is faster and more energy efficient in multi-task, multi-server scenarios. Using the DPTOS approach results in a reduction in the total energy consumption of UE by 38.9%, 23.41%, 18.62%, and 13.72% compared to the EMMOS, RPTOS, TOOS, and EEDOS benchmark approaches, respectively, with 74.2% accuracy of the trained deep neural network.