From an end-to-end performance perspective, a multi-agent-based cognitive resource management (MA-CRM) framework is proposed in this article. More importantly, we introduce a novel concept of resource flow (RF) as integrated CRM in a multi-dimensional environment scenario. We concentrate on the autonomous CRM framework in cognitive radio (CR) networks. Specifically, we first summarize the necessity of the novel concept for implementing intelligent and autonomous resource management considering both the requirements of various types of resource and the unified framework of the CRM scheme. Then, we introduce the concept of RF, including the technical aspects, purpose, classification, and description. Finally, we give a use case of RF for autonomous CRM, where the optimal RF is achieved to guarantee the resource-imbalanced requests of different service traffic flows. With the rapid growth in mobile communications, deployment and maintenance of future networks are becoming more and more complicated, time-consuming, and expensive [1][2][3]. In most existing networks, parameters are mutually adjusted to achieve a high level of network performance. In Long Term Evolution (LTE), the concept of a self-organizing network is introduced, where system parameter tuning is done astronomically based on network measurements [4]. Meanwhile, cognitive terminals in future convergence networks of multi-heterogeneous radio access networks will be faced with a flexible environment including radio, networks, and user context. Therefore, more efficient cognitive resource management (CRM) strategies and algorithms have to be integrated in future mobile networks to further reduce capital expenditure (CAPEX) and operational expenditure (OPEX) [5][6][7][8][9]. In addition, the deployment and optimization of mobile networks are significantly complicated and challenging engineering tasks requiring a comprehensive systematic CRM and allocation approach, which determines the overall performance of the current communication systems and better guarantees the varying quality of service (QoS) of the different traffic flows [1,7,9]. Under the condition of high convergence of heterogeneous networks, how to achieve effective utilization of multidomain resources is a great challenge and filled with strategic meaning. Highly efficient use of resources not only means more revenue for the operators, but also bridges the gap between resource scarcity and resource under-utilization. To break the resource gridlock [10], CRM has been followed with great interest. For instance, multi-agent system-based cognitive resource management architecture is proposed in [11], while a multi-agent learning optimal power control policy and channel selection without negotiation in CR systems are investigated respectively in [11] and [12]. In [13], layer-less dynamic networks enabling the characterization of fundamental performance regions without a presupposition of protocol stack layering are considered. This study is typically distinguished from previous results of cross-...