CS-171, Intro to A.I.

— Final Exam — Fall Quarter, 2012

YOUR ID: ID TO RIGHT: ROW: NO. FROM RIGHT:

The exam will begin on the next page. Please, do not turn the page until told.

When you are told to begin the exam, please check first to make sure that you
have all 10 pages, as numbered 1-10 in the bottom-left corner of each page.

The exam is closed-notes, closed-book. No calculators, cell phones, electronics.

Please clear your desk entirely, except for pen, pencil, eraser, an optional blank
piece of paper (for optional scratch pad use), and an optional water bottle.
Please turn off all cell phones now.

This page summarizes the points available for each question so you can plan your time.

1. (10 pts total) Decision Tree Classifier Learning.

2. (5 pts total, -1 pt each wrong answer, but not negative) Search Properties.

3. (10 pts total) Naïve Bayes Classifier Learning.

4. (15 pts total, 5 pts each, -1 each error, but not negative) Bayesian Networks.

5. (10 points total, 2 pts each) Constraint Satisfaction Problems.

6. (10 pts total, -1 for each error, but not negative) Alpha-Beta Pruning.

7. (10 pts total, -2 for each error, but not negative) Conversion to CNF.

8. (10 pts total, -2 for each error, but not negative) Resolution Theorem Proving.

9. (10 pts total, 1 pt each) State-Space Search.

10. (10 pts total, 2 pts each) English to FOL Conversion.

The Exam is printed on both sides to save trees! Work both sides of each page!

1
2
1. (10 pts total) Decision Tree Classifier Learning. You are a robot in a lumber yard,
and must learn to discriminate Oak wood from Pine wood. You choose to learn a
Decision Tree classifier. You are given the following examples:
Example Density Grain Hardness Class
Example #1 Heavy Small Hard Oak
Example #2 Heavy Large Hard Oak
Example #3 Heavy Small Hard Oak
Example #4 Light Large Soft Oak
Example #5 Light Large Hard Pine
Example #6 Heavy Small Soft Pine
Example #7 Heavy Large Soft Pine
Example #8 Heavy Small Soft Pine
1a. (2 pts) Which attribute would information gain choose as the root of the tree?

1b. (4 pts) Draw the decision tree that would be constructed by recursively applying
information gain to select roots of sub-trees, as in the Decision-Tree-Learning algorithm.

Classify these new examples as Oak or Pine using your decision tree above.
1c. (2 pts) What class is [Density=Light, Grain=Small, Hardness=Hard]?
1d. (2 pts) What class is [Density=Light, Grain=Small, Hardness=Soft]?

2. (5 pts total, -1 pt each wrong answer, but not negative) Search Properties.
Fill in the values of the four evaluation criteria for each search strategy shown. Assume
a tree search where b is the finite branching factor; d is the depth to the shallowest goal
node; m is the maximum depth of the search tree; C* is the cost of the optimal solution;
step costs are identical and equal to some positive ε; and in Bidirectional search both
directions use breadth-first search.
Note that these conditions satisfy all of the footnotes of Fig. 3.21 in your book.
Criterion Complete? Time complexity Space complexity Optimal?
Breadth-First
Uniform-Cost
Depth-First
Iterative Deepening
Bidirectional
(if applicable)

3
3. (10 pts total) Naïve Bayes Classifier Learning. You are a robot in a lumber yard,
and must learn to discriminate Oak wood from Pine wood. You choose to learn a Naïve
Bayes classifier. You are given the following (noisy) examples:
Example Density Grain Hardness Class
Example #1 Light Small Hard Oak
Example #2 Heavy Large Hard Oak
Example #3 Heavy Small Soft Oak
Example #4 Heavy Small Soft Oak
Example #5 Light Large Hard Pine
Example #6 Light Small Soft Pine
Example #7 Heavy Large Soft Pine
Example #8 Light Large Hard Pine

Recall that Baye’s rule allows you to rewrite the conditional probability of the class given
the attributes as the conditional probability of the attributes given the class. As usual, α
is a normalizing constant that makes the probabilities sum to one.

P(Class | Density, Grain, Hardness) = α P(Density, Grain, Hardness | Class) P(Class)

3a. (2 pts) Now assume that the attributes (Density, Grain, and Hardness) are conditionally
independent given the Class. Rewrite the expression above, using this assumption of
conditional independence (i.e., rewrite it as a Naïve Bayes Classifier expression).

3b. (4 pts total; -1 for each wrong answer, but not negative) Fill in numerical values
for the following expressions. Leave your answers as common fractions (e.g., 1/4, 3/5).

P(Oak)= P(Pine)=

P(Density=Light | Class=Oak)= P(Density=Light | Class=Pine)=

P(Density=Heavy | Class=Oak)= P(Density=Heavy | Class=Pine)=

P(Grain=Small | Class=Oak)= P(Grain=Small | Class=Pine)=

P(Grain=Large | Class=Oak)= P(Grain=Large | Class=Pine)=

P(Hardness=Hard | Class=Oak)= P(Hardness=Hard | Class=Pine)=

P(Hardness=Soft | Class=Oak)= P(Hardness=Soft | Class=Pine)=

P(Class=Oak | Density=Heavy ^ Grain=Small ^ Hardness=Hard) =

P(Class=Pine | Density=Heavy ^ Grain=Small ^ Hardness=Hard) =

4
4. (15 pts total, 5 pts each, -1 each error, but not negative) Bayesian Networks.
4a. (5 pts) Draw the Bayesian Network that corresponds to this conditional probability:

P(A | C,D,F) .P(B | D,E) P(C | F) P(D | G) P(E | G) P(F | H) P(G | H) P(H)

4b. (5 pts) Write down the factored conditional probability expression that corresponds
to the graphical Bayesian Network shown.

G H

D E F

B C

A
4.c. (5 pts) Shown below is the Bayesian network corresponding to the Burglar Alarm problem,
P(J | A) P(M | A) P(A | B, E) P(B) P(E).
P(E) A P(M) B E P(A)
(Alarm) .002 t .70 t t .95
(Burglary) B E (Earthquake) f .01 t f .94
A A P(J) f t .29
(Mary calls) P(B) t .90 f f .001
(John calls) J M .001 f .05

The probability tables show the probability that variable is True, e.g., P(M) means P(M=t).
Write down an expression that will evaluate to P( j=t ∧ m=f ∧ a=f ∧ b = f ∧ e = t). Express your answer as
a series of numbers (numerical probabilities) separated by multiplication symbols. You do not need to
carry out the multiplication to produce a single number (probability). SHOW YOUR WORK.

5
5. (10 points total, 2 pts each) Constraint Satisfaction Problems.

VT NH ME
CT

RI MA

You are a map-coloring robot assigned to color this New England USA map. Adjacent regions
must be colored a different color (R=Red, B=Blue, G=Green). The constraint graph is shown.

5a. (2pts total, -1 each wrong answer, but not negative) FORWARD CHECKING.
Cross out all values that would be eliminated by Forward Checking, after variable MA
has just been assigned value R as shown:
CT RI MA VT NH ME
RGB RGB R RGB RGB RGB

5b. (2pts total, -1 each wrong answer, but not negative) ARC CONSISTENCY.
CT and RI have been assigned values, but no constraint propagation has been done.
Cross out all values that would be eliminated by Arc Consistency (AC-3 in your book).
CT RI MA VT NH ME
R G RGB RGB RGB RGB

5c. (2pts total, -1 each wrong answer, but not negative) MINIMUM-REMAINING-
VALUES HEURISTIC. Consider the assignment below. RI is assigned and constraint
propagation has been done. List all unassigned variables that might be selected by the
Minimum-Remaining-Values (MRV) Heuristic: .

CT RI MA VT NH ME
RB G RB RGB RGB RGB

5d. (2pts total, -1 each wrong answer, but not negative) DEGREE HEURISTIC.
Consider the assignment below. (It is the same assignment as in problem 5c above.) RI
is assigned and constraint propagation has been done. List all unassigned variables
that might be selected by the Degree Heuristic:. .

CT RI MA VT NH ME
RB G RB RB RGB RGB

5e. (2pts total) MIN-CONFLICTS HEURISTIC. Consider the complete but inconsistent
assignment below. MA has just been selected to be assigned a new value during local
search for a complete and consistent assignment. What new value would be chosen
below for MA by the Min-Conflicts Heuristic?. .

CT RI MA VT NH ME
B G ? G G B

6
6. (10 pts total, -1 for each error, but not negative) Alpha-Beta Pruning. In the
game tree below it is Max's turn to move. At each leaf node is the estimated score of
that resulting position as returned by the heuristic static evaluator.
(1) Perform Mini-Max search and label each branch node with its value.
(2) Cross out each leaf node that would be pruned by alpha-beta pruning.
(3) What is Max’s best move (A, B, or C)?

(Max)

(Min)

(A) (B) (C)

(Max)

4 3 8 2 2 2 4 1 2 8 5 2 3 4 9 5 1 8 3 9 4 3 2 4 1 1 2

7. (10 pts total, -2 for each error, but not negative) Conversion to CNF. Convert this
Propositional Logic wff (well-formed formula) to Conjunctive Normal Form and simplify.
Show your work (correct result, 0 pts; correct work, 10 pts).

[¬(Q⇒P)]⇔P

7
8. (10 pts total, -2 for each error, but not negative) Resolution Theorem Proving. You are a
robot in a logic-based question answering system, and must decide whether or not an input goal
sentence is entailed by your Knowledge Base (KB). Your current KB in CNF is:

S1: ( P Q )
S2: ( ¬P Q )
S3: ( P ¬Q )
S4: ( ¬P R )

Your input goal sentence is: ( P ∧ Q ∧ R).

8a. (2 pts) Write the negated goal sentence in CNF.

S5:

8b. (8 pts total, -2 for each error, but not negative) Use resolution to prove that the goal
sentence is entailed by KB, or else explain why no such proof is possible. For each step of the
proof, fill in Si and Sj with the sentence numbers of previous CNF sentences that resolve to
produce the CNF result that you write in the resolvent blank. The resolvent is the result of
resolving the two sentences Si and Sj. Use as many steps as necessary, ending by producing
the empty clause; or else explain why no such proof is possible.

Resolve Si with Sj to produce resolvent S6:

Resolve Si with Sj to produce resolvent S7:

Resolve Si with Sj to produce resolvent S8:

Resolve Si with Sj to produce resolvent S9:

Resolve Si with Sj to produce resolvent S10:

Resolve Si with Sj to produce resolvent S10:

Add additional lines below if needed; or, if no such resolution proof is possible, use the space
below to explain why not:

8
9. (10 pts total, 1 pt each) State-Space Search. Execute Tree Search through this graph (do
not remember visited nodes, so repeated nodes are possible). It is not a tree, but pretend you
don’t know that. Step costs are given next to each arc, and heuristic values are given next to
each node (as h=x). The successors of each node are indicated by the arrows out of that node.
(Note: D is a successor of itself). As usual, successors are returned in left-to-right order.
The start node is S and the goal node is G. For each search strategy below, indicate
(1) the order in which nodes are expanded, and (2) the path to the goal that was found, if any.
Write “None” for the path if the goal was not found. The first one is done for you, as an example.

S h=21

4
10
h=18
3 h=15 10
A B h=5

9 6 6
D h=16 G
9.a. DEPTH-FIRST SEARCH:

9.a.(1) Order of expansion: S A B D D D D ...

9.a.(2) Path to goal found: None

9.b.(1) Order of expansion:

9.b.(2) Path to goal found:

9.c.(1) Order of expansion:

9.c.(2) Path to goal found:

9.d.(1) Order of expansion:

9.d.(2) Path to goal found:

9.e.(1) Order of expansion:

9.e.(2) Path to goal found:

9.f.(1) Order of expansion:

9.f.(2) Path to goal found:

9
10. (10 pts total, 2 pts each) English to FOL Conversion. For each English sentence
below, write the FOL sentence that best expresses its intended meaning. Use Dog(x)
for “x is a dog,” Bone(x) for “x is bone,” and Likes(x, y) for “x likes y.”
The first one is done for you as an example.

10a. (2 pts) “Every dog likes every bone.”

∀x ∀y [ Dog(x) ∧ Bone(y) ] ⇒ Likes(x, y)

10b. (2 pts) “Some dog likes some bone.”

10c. (2 pts) “For every dog, there is a bone that the dog likes.”

10d. (2 pts) “For every bone, there is a dog who likes that bone.”

10e. (2 pts) “There is a bone that every dog likes.”

10f. (2 pts) “There is a dog who likes every bone.”

10