The demand for mobile services is increasing constantly and mobile network operators need to significantly upgrade their networks to respond to the demand. The increasing complexity of the networks makes it impossible for a human operator to manage them optimally. Currently the network management operations are automated using a pre-defined logic. The future goal is to introduce cognitive network management functions which can adapt to changes in the network context and handle uncertainty in network data.
This thesis discusses using Markov Logic Networks for cognitive management of mobile networks. The method allows uncertain and partial information and makes it possible to consolidate knowledge from multiple sources into a single, compact, representation. The model can be used to infer configuration changes in network parameters and the model parameters can be learned from data.
We test the method in a simulated LTE network and examine the results in terms of improvements in network performance and computational cost.
