Midterm

Your paper must cover one of the following topics:

• Summarise, compare and contrast the various techniques for solving Markov Decision Prcesses that we covered in the first six lectures (Dynamic Programming, Monte-Carlo approaches and Temporal Difference learning).

  To get full credit you must not only describe these techniques but also explain how they relate to one another, and suggest when each of them would be the most appropriate technique to adopt.

• Provide a critical discussion of two of the papers that we have covered in the last six lectures.

  For full credit you must not only briefly describe the work in the papers, but also discuss the good and bad features of the papers (justifying your position).

• Describe how reinforcement learning techniques might be applied to some problem of interest to you.

  To get full credit you must describe the problem in detail (enough detail that I can understand it :-), and also describe how the relevant components of the reinforcement learning model (the reward structure, the probability distributions and so on) may be obtained for the problem.

You need to hand in your paper at the class on March 30th.