The key theme of reinforcement learning is that it is a computational approach of dynamic optimization modeling. Due to mathematical requirements understanding dynamic optimization, the first half of the course will be split to Science and Business tracks.

In the Science track, students are given introduction of advanced optimization techniques such as Bellman equation and Hamiltonian approaches. On the contrary, the business track students are to extend infinite serial models from earlier courses into toy version of business applications in dynamic optimization.

In the second half, although the problem sets will be varying between two tracks, the class topics will be common, which are bullet pointed below

Q-learning

• Value-based
  – Experience replay
• Policy-based
  – REINFORCE algorithm
  – Actor-Critic algorithm
  – other variance reduction algorithms

Game theory in dynamic optimization

• Robot Football – Cooperation vs. Competition
• Self-driving automobile
• Defensive driving

Due to technical limitation, the depth of analysis on above topic will not be advanced. More technically advanced approaches will be discussed in Advanced Reinforcement Learning