Abstract

IoT (Internet of Things) and cloud computing are essential components of constructing smart cities and provide multiple smart services to end consumers. Because IoT devices are data-intensive and resource-constrained, using edge computing technologies might provide considerable benefits to the smart environment. However, heterogeneous clouds where ECSs (Edge Cloud Systems) and centralized clouds interact for satisfying demands of IoT applications, challenges in task offloading exist. When IoT system's environments change, such as the edge server's performances or the bandwidth, solutions based on DLTs (Deep Learning Techniques) must train from scratch. A meta-algorithm known as DDMTO (Distributed Deep Meta learning-driven Task Offloading) is presented to solve the issue of poor portability and ensure that DNNs (Deep Neural Networks) are utilised to make offloading decisions effectively and efficiently. These networks have their output which receives inputs from hidden layers in BP algorithm compute outputs. Inputs are compared to desired outputs and errors are traced from outputs to hidden layers and from hidden to input layers based on disparities. When flows are restored, neuron weights get altered. Epochs are cycles that traverse from inputs to outputs and backwards from outputs to inputs. Previously known inputs are fed into NNs (neural networks) which then generate known outputs called network training. Existing offloading systems ignore heterogeneous cloud co-operations which is overcome for providing better performances while significantly reducing computing complexities.