This module focuses on the interplay between learning (in particular reinforncement learning) and digital communications. This module covers Markov Decision Process (MDP) and partially observable MDP as the foundation of reinforcement lerning (RL). You will also learn about using neural networks to obtain the optimal action (control) policy. Distributed RL and its relationship with distributed optimization will also be discussed. In this module, we will use communication system design and control to illustrate the applicability of RL.

By the end of the module, you will be able to: - model complex systems using Markov Decision Process (MDP) and partially observable MDP, - critique the components of RL including value function approximation, temporal difference learning, SARSA, Q-learning, policy gradient methods, exploration and exploitation, - implement deep RL using neural networks, - Design RL algorithms for different applications, - apply RL to communication systems and networks.

The module will cover: (1) Markov Decision Process (MDP) and partially observable MDP as the foundation of reinforcement learning (RL), (2) value function approximation, temporal difference learning, SARSA, Q-learning, policy gradient methods, exploration and exploitation, (3) deep RL using neural networks to obtain the optimal action (control) policy, (4) established RL algorithms in the literature such as DQN, (5) Distributed RL and its relationship with distributed optimization, and (6) application of RL to the design and control of communication systems and networks.