CS 540: Introduction to Artificial Intelligence

Final Exam: 12:25-2:25pm, May 16, 2013

Room 228 Educational Sciences

CLOSED BOOK
(two sheets of notes and a calculator allowed)

Write your answers on these pages and show your work. If you feel that a question is not fully
specified, state any assumptions that you need to make in order to solve the problem. You may
use the backs of these sheets for scratch work.

Write your name on this page and initial all other pages of this exam. Make sure your exam
contains eight problems on ten pages.

Name ________________________________________________________________

Student ID ________________________________________________________________

Problem Score Max Score___

1 ______ 15

2 ______ 16

3 ______ 15

4 ______ 15

5 ______ 10

6 ______ 10

7 ______ 10

8 ______ 9

TOTAL ______ 100

 Initials: ________________________

Problem 1 – Bayesian Networks (15 points)

Consider the following Bayesian Network, where variables A-D are all Boolean-valued:
A P(B =true | A)
P(A=true) = 0.7 A B
false 0.9
true 0.2

A P(C =true | A) C
B C P(D =true | B,C)
false 0.6 D false false 0.4
true 0.3
false true 0.5
true false 0.8
true true 0.1

a) What is the probability that A and B are true but C and D are false? ______________
[Be sure to show your work for Parts a-c.]

b) What is the probability that B is true, A is false, and D is true? _______________

c) What is the probability that B is true given that A is false and D is true? _______________

2
 Initials: ________________________

Problem 2 – Probabilistic Reasoning and Learning (16 points)

a) Consider the following training set, where all possible values of the three features appear in
it. Assume you wish to apply the Naïve Bayes algorithm.

Ex # A B C Output
1 2 Small Red True
2 1 Small Green True
3 3 Large Blue False
4 2 Small Blue False
5 4 Large Green False

Calculate the ratio below, showing your work below it and putting your (numeric) answer on
the line to the right of the equal sign. Be sure to explain how you performed Laplace
smoothing.

Prob(Output = True | A = 2, B = Large, C = Green)

__________________________________________________ = ___________

Prob(Output = False | A = 2, B = Large, C = Green)

What is the most likely output for this example (A=2, B=Large, and C=Green)? _________
Briefly explain your answer below.

3
 Initials: ________________________

b) Assume we wish to use a full joint probability table for this task.

How many cells are needed if represented as a single multi-dimensional Java array?
Briefly explain your answer.
Answer: ______________

Draw the full joint probability table as a Bayes Net. Show the conditional probability
table associated with each node, but you do not need to fill the cells with actual numbers
(i.e., just leave them blank).

Here again is the dataset from Part a, copied for your convenience:

Ex # A B C Output
1 2 Small Red True
2 1 Small Green True
3 3 Large Blue False
4 2 Small Blue False
5 4 Large Green False

4
 Initials: ________________________

Problem 3 – Representing Knowledge with First-Order Logic (15 points)

Convert each of the following English sentences into First-Order Predicate Calculus (FOPC),
using reasonably named predicates, functions, and constants. If you feel a sentence is
ambiguous, clarify which meaning you’re representing in logic. (Write your answers in the
space below each English sentence.)

Colin is a person who knows someone who owns a red bike.

For every species of tree in the world there is a bacterium that only lives in this type of tree.

Usually children are younger than their step parents, but not always.
[You must use the notation of Markov Logic Networks here.]

5
 Initials: ________________________

Problem 4 – Logical Reasoning (15 points)

a) Is the following wff valid, satisfiable, or unsatisfiable (circle one)? Justify your answer
formally in the space below.

[ (P  Q) ˅ (Q  P) ]  (P ↔ Q)

b) Put this wff in clausal form; show your work below and put your answer on the line.

¬  x [ p(x)  q(x) ] ____________________________________

c) We wish to use search to find an interpretation (i.e., a ‘world state’) that satisfies all the
clauses below. Show how this set of clauses can be simplified before starting the search.
Justify your simplification(s).

(Q  A  ¬ R)

(P  ¬ R)

(P  ¬B  ¬ Q)

(¬ P  R)

(A  B)

6
 Initials: ________________________

Problem 5 – More Logical Reasoning (10 points)

a) What is the most-general unifier (mgu), if any, of these two wff’s? ___________________

Q(?x, ?y, ?x) Q(2, f(1,?z, 2), ?z)

b) Given the following background knowledge:

1.  x { p(x)  [ q(x)  r(x) ] }

2.  y [ q(y)  s(y) ]
3. p(1)
4. p(2)
5. r(1)
6.  r(2)

Show: z s(z) by filling out the table below, using as many lines as needed.

Number WFF Justification

7
_____________________________________________________________________________________________
8
_____________________________________________________________________________________________
9
_____________________________________________________________________________________________
10
_____________________________________________________________________________________________
11
_____________________________________________________________________________________________
12
_____________________________________________________________________________________________
13
_____________________________________________________________________________________________
14
_____________________________________________________________________________________________
15

7
 Initials: ________________________

Problem 6 – Artificial Neural Networks (10 points)

Consider a perceptron that has two real-valued inputs and an output unit that uses a step function
as its output function. All the initial weights and the output unit’s threshold equal 0.05. Assume
the teacher has said that the output should be 1 for the input in1 = 0.2 and in2 = -0.7.

a) Show how the perceptron learning rule (also called the delta rule) would alter this neural
network upon processing this training example. Let  (the learning rate) be 0.3.

Perceptron BEFORE Training

Perceptron AFTER Training

b) Did these changes improve the network? _________ Explain your answer below.

c) What kind of search does the perceptron learning rule perform? ______________ Explain.

8
 Initials: ________________________

Problem 7 – Support Vector Machines (10 points)

a) Draw and explain a picture that illustrates the idea and value of searching for a large margin.
Be sure to compare (verbally and in your picture) to what the perceptron learning rule,
without a “weight decay” term, might do.

b) Consider the dataset below. Assume we wish to use the idea of ‘kernels’ to convert this
dataset into a new dataset, one where we will apply the perceptron learning rule (this time,
with weight decay). Our chosen similarity function is below. Show the new dataset below
the old one and be sure to clearly label the columns and rows.

The number of feature pairs (A+B, B+C, and C+A) that have the same values in the two
examples (e.g., since ex1 and ex3 both have the same values for features A and B, that counts as
one, whereas ex1 and ex4 do not have the same values for the feature pair A+B and nothing is
added to the count).

Ex # A B C Output
1 True False False True
2 False True True True
3 True False True False
4 True True False False

9
 Initials: ________________________

Problem 8 – Important AI Concepts (9 points)

Describe each of the following AI concepts and briefly explain its most significant aspect.
(Write your answers in the space below the AI concept.)

CLOSED List
description:

significance:

Overfitting

description:

significance:

Strong AI
description:

significance:

Have a good summer vacation!

10