—Policy Based SON Management (PBSM) framework has been introduced to manage Self-Organizing Networks (SON) functions in a way that they fulfill all together the operator global goals and provide a unique self-organized network that can be controlled as a whole. This framework mainly translates operator global objectives into policies to be followed by individual SON functions. To cope with the complexity of radio networks due to the impact of radio environment and traffic dynamics , we propose to empower the PBSM with cognition capability. We propose a Cognitive PBSM (CPBSM) that relies on a Reinforcement Learning (RL) algorithm which learns the optimal configuration of SON functions and steers them towards the global operator objectives. The visitor will see how changing the operator objectives leads to a reconfiguration of the SON functions in such a way that the new objectives are fulfilled. The operation of a RL based cognitive management will be illustrated and the exploration/exploitation and scalability dilemmas will be explained.