What will be in the CPSC 422 Midterm (24 February 2005)
****************************************************
We will have some subset (not necessarily strict subset) of the
following questions:

Question 1 (robot control)

I will select one or two of the following questions (perhaps reworded). Use
proper English. Be concise in your answers. You must use your own
words (text from the textbook or another source copied onto your crib
sheet will not get any marks).

(a) Explain why we use hierarchical (layered) controllers.

(b) What does it mean that the higher layers run at a different time
scale than lower layers?

(c) Give the intuition behind the notion of a transduction? Why do we
define causal transductions?

(d) Why does an agent have a state?

(e) Explain why we need a command function and a state transition
function in a robot controller but not other functions.

(f) Explain how the communication between layers and communication between
time steps is handled in the logical implementation of a hierarchical
controller. 

(g) Why don't we run the logical specification of a hierarchical
controller using SLD resolution? How can it be implemented
efficiently?

Question 2 (decision-theoretic planning)

One or more of the following:

Show how to do one step of value iteration.

Explain the advantages and disadvantages of storing the Q function as
compared to storing the Value function.

Explain how decision-theoretic planning fits into the robot control
architecture. How can a robot controller be built using
decision-theoretic planning.

Question 3 (reinforcement learning)

One or more of the following:

Show how to do one step of Q-learning, or the effect of a sequence of steps.

Explain why SARSA(lambda) learns faster than Q-learning.

Explain why alpha_k should be reduced as function of k. Explain why
you may not want to reduce alpha.

Explain how Q-learning fits into the robot control architecture. How
can a robot controller be built using Q-learning?

There are a number of parameters in SARSA-lambda: alpha, gamma,
lambda. Explain what each does. Which one affects what is the correct
answer? Which one affects whether it will converge?

Question 4 (Assumption-based reasoning).

One or more of the following:

Find the conflicts and the consistency-based diagnoses for a
particular example domain, (e.g., Exercise 7.1, or perhaps a more
realistic example).

find the abductive diagnoses or model a domain so you can find the
abductive diagnoses. (e.g., exercise 9.1).

a question about default prediction or about the combination of
default and abductive reasoning. (What you should predict
based on normality assumptions).

Give a meta-interpreter for a particular example.

Question 5 (Assignment)

Some questions about what you should have learned from doing
assignments. For example:

Show how one step of value iteration, Q-learning, SARSA-lambda works
for the game domain of assignment 2.