As network infrastructure moves towards virtualized infrastructure, Operations Support Systems (OSS) and Business Support Systems (BSS) are evolving in the directions of virtualization based on different types of services such as IaaS (Infrastructure as a Service), PaaS (Platform as a Service ) or SaaS (software as a service). This development involves the use of shared commercial hardware and storage, enabling the provision of a separate virtualized infrastructure, virtualized platform, and virtualized software to a group of users using a shared physical infrastructure. Specific application environments can then be made available for various implementations including artificial intelligence and machine learning capabilities to provide the advanced analytics needed to rapidly bring services to market [O39].

ML and DL techniques are data-driven and require quality data. This data may not be available and is extremely difficult to obtain. For example in the case of a DL-based resource management method that addresses the problem of channel allocation and performance in a large cellular IoT network, for a given network state (e.g., the number of users, their affiliation to a BS, channel gains to the respective BS) will require data to train the DNN on optimal channel allocation and performance. Generating this data will be very computationally expensive, even using techniques such as genetic algorithms. Therefore, data generation issues will need to be addressed in order to develop DL-based resource management schemes.