CS7642_Readings

Question 1

Any finite length discrete time MDP can be converted to a TTD-MDP? (True/False)

Answer 1

True. This is from the TTD MDP paper for week 10

Question 2

It is always possible to find a stochastic policy that always solves a TTD-MDP?

Answer 2

False. See Bhat 2007

Question 3

In policy shaping, probably the most important parameter is gamma?

Answer 3

False. The most important parameter for learning in policy shaping is the probability that an evaluation of an action choice is correct for a given learner (denoted by ‘C’)

Question 4

An Oracle is always better than human teaching agents in policy shaping?

Answer 4

False. Cederbourg 2011 says this is not the case

Question 5

Like MDPs, every Markov game has a non-empty set of optimal policies, at least one of which is stationary? (True/False)

Answer 5

True. From Littman 1994

Question 6

There is always at least one optimal stationary deterministic policy for Markov games? (True/False)

Answer 6

False. Only regular MDPs make this guarantee.