Due the extensive proliferation of wireless applications and services over the past decade, the utilization of wireless network resources has increased, which subsequently led to network resources shortage. In this respect, cognitive radio network (CRNs) has envisaged as a promising technology to utilize the network resources efficiently. Recently, the cognitive paradigm was extended to network management allowing the flexible allocation of network resources based on the concepts of virtualization and softwerization. However, current cognitive communication approaches suffer from two major shortcomings: 1) they are often based on statistical modelling without exploiting data-aided methods based on Machine Learning (ML), 2) they focus on a specific part of the system without addressing the cross-layer aspects. Hence, in this project, we propose Learning-Assisted Cross-Layer Optimization framework for Cognitive Communication Networks (LACLOCCN), the main objective is to exploit ML techniques to actively acquire side information and iteratively optimize the system resources across all communication layers.