Meta-Learning for Realizing Self-x Management of Future Networks

In this paper, we propose an autonomic network management and policy execution framework. The proposed framework refactors the network functionalities by decomposing the network architecture into hierarchical layered architecture. This paper aims at enabling the transition from a rule-based control structure to a more distributed and autonomic network control by implementing the self-x or self-* learning vision on each layer. The problem is modeled using multi-layer dynamic games. At each layer, a self-* learning procedure is proposed to learn and adapt the reverse Stackelberg policies. To validate the proposed framework, we develop a full scale demonstrator comprising of flat IP core and heterogeneous wireless access networks. We have also developed various tools and software agents to implement the self-x management vision. The proposed self-x learning is implemented via mobile intelligent agents in a distributed fashion. Our experimental results show quick re-stabilization of the self-* learning in mobile intelligent agents and the observed performance remain well above the satisfactory values for different key performance indicator with the proposed meta-learning approach.