Multiresource allocation in dynamic MEC systems: a continual learning-based approach

Recently, mobile edge computing (MEC) has been considered as a promising solution to address the challenges arising from the growing computationally intensive applications and limited computation capability of mobile devices. Thus, the resource allocation is crucial for ensuring the system performance of MEC. Recent deep learning-based resource allocation (DLRA) methods achieve preferable performance in real-time. However, in practical scenarios, the number of users often changes, which leads to a decrease in the performance of DLRA methods. Thus, once the number of users changes, the traditional DLRA methods need to retrain their neural networks, which consumes a lot of computing resources. To address this issue, this paper proposes a continual learning-based resource allocation (CLRA) method to jointly optimize multiple resource including offloading decision, power control, time slot assignment and computation resource distribution in MEC system. Our CLRA method is able to avoid forgetting previous knowledge when the number of users in the MEC system changes by selecting a small number of the most representative samples from the past and relearning. Experimental results show that, our proposed CLRA method achieves a better performance than the traditional DLRA method in terms of the higher quality of service's (QoS's) satisfaction rate and the less energy consumption.