Application of machine learning in communication networks

Abstract

Human curiosity has always led to the development of technologies. To mimic the hu man brain in machines, led to the development of Artificial Intelligence (AI). The aim is to develop a general AI which can learn multiple complex tasks and make decisions. Major work that has been done till now falls in the category of artificial narrow intel ligence (ANI), where multiple single tasks have been mastered like object detection, speech to text, predictions, etc. All this is made possible by machine learning (ML) and deep learning (DL) techniques that learn from huge labeled data source which is abundantly available in this era. Reinforcement learning (RL) is a different learning technique than the previous ML methods. It is an experience based learning which uses rewards and punishment to build an optimal policy, just like a human baby learns. Combining DL and RL techniques to gether form a very powerful method called deep reinforcement learning (DRL). This is the initial step in the artificial general intelligence (AGI) category. In this thesis we have explored the DRL technique for optimization of routing and resource allocation in quantum key distribution (QKD) networks in optical networks. DRL has been successful in serving the purpose in simulated environments, thus im prove the resource utilization and reduce the blocking of requests in the QKD network as compared to the baseline algorithms.