On Applicability of Imagery-Based CNN to Computational Offloading Location Selection

The progress in computational offloading is heavily pushing the development of modern Information and Communications Technology domain. growth resource-constrained Internet Things devices demands new strategies to be sustainably integrated beyond 5G networks. One solutions said demand enabling Mobile Edge Computing (MEC) powered by advanced methods Machine Learning (ML). This paper proposes application ML-powered strategy a wireless cellular network applying traditional fundamental Travelling Salesman Problem (TSP) on location selection. main specificity proposed approach use imagery data. Thus, executes literature review identify existing strategies. It further novel method utilizing location-like data most suitable executing search for an identified route between locations using Deep (DL) model. model was evaluated achieved MAE – 1,575, MSA 10,119,205, $R^{2}$ 0.98 testing dataset, which outperforms or comparable with other well-known architectures. Moreover, training time proven 2–10 times faster. Interestingly, values are relatively low compared target that should predicted (despite rather high MSE results), confirmed almost perfect value. concluded neural can predict values, this solution applied real-world tasks.