CSE 352 ARTIFICIAL INTELLIGENCE

1
CSE 352 ARTIFICIAL INTELLIGENCE
Summer 2010 Take Home Final Test
Due: July 8 in class.
There are 3 pages in this exam. Please make sure you have all of them
INTRODUCTION – Philosophical AI Questions (10pts)
Q1. Write some pro and contra arguments about TURING TEST as the test of artificial
intelligence.
Q2. Describe some technological “science fiction” inventions in your favorite Science
Fiction movie that are realistic as of today’s state of technology, and some others that
are not realistic.
PART ONE –Expert Systems (15pts)
Q1. Write a simple (3-4 rules long) Expert System (in English) for a task (or tasks) chosen
by you.
Q2. Conceptualize your system in PROPOSITIONAL LOGIC.
Q3. Pose some questions to your system in order to build a database of FACTS related to
your task and use backward chaining to show if your task succeeds, or fails depending on the
answers to the questions.
HINT: Look at Hmk1part1 solutions and our book.
PART TWO –Predicate Logic (25pts)
Q1. Prove the following Law of Quantifiers
( x A( x)  x B ( x))
x ( A( x)  B ( x))
Give an example of a non empty domain X and formulas A(x), B(x),
For which the
direction does not hold.
Q2. (a) Translate the following sentences into Predicate Logic and AI Logic under
intended interpretation. Write down your choice of predicates.
A1: All flowers are blue and pretty.
A2: All blue flowers are pretty.
A3: All flowers are pretty.
2
(b) Decide whether the ARGUMENT: A1 /\ A2 => A3 is VALID, or NOT VALID.
Show work.
Q6. (a) Translate the following sentences into Predicate Logic and AI Logic under
intended interpretation. Write down your choice of predicates.
A1: All students are clever.
A2: Some clever students have blue eyes.
A3: There is a student with blue eyes.
(b) Decide whether the ARGUMENT: A1 /\ A2 => A3 is VALID, or NOT VALID.
Show work.
PART THREE – Classification by Decision Tree Induction (30pts)
Given a TRAINING and TEST classification data as follows:
TRAIN
Record
o1
o2
o3
o4
o5
o6
o7
o8
a1
1
0
0
0
1
1
0
0
a2
1
0
1
0
1
1
0
0
a3
0
1
1
1
0
1
0
1
C
+
+
+
-
TEST
Record
o1
o2
o3
o4
a1
1
1
0
0
a2
1
0
0
0
a3
0
1
1
0
C
+
+
-
Q1. Use attribute a2 as the ROOT of a Decision Tree. Construct the tree (show steps) using
your own choice of all node attributes. NO general majority Voting, i.e. must use
Algorithm 1
Q2. Write down all the discriminant rules determined by your tree in two forms:
1. using att = v expression and
2. in a PREDICATE form using expressions
att(x, v) instead of att = v.
3. Evaluate Accuracy of your rules.
Q3. Use the TEST data to evaluate predictive accuracy of your set of rules generated by
the given TRAIN data. Specify the testing method you use.
3
Q4. Re-write your TEST data using your choice of “real” values for attributes and values
instead of our “abstract” attributes values.
Example
a1 = age,
a2 = color,
1 = old,
1 = blue,
0 = young
0 = red
PART FOUR– Learning Neural Networks (20pts)
Given two records (Training Sample)
a1
0.5
0
a2
0
0.3
a3
0.2
0
Class
1
1
Use the Network below to evaluate a passage of TWO EPOCHS.
1
w14
w15
2
4
w46
w24
6
w25
w34
3
5
w56
w35
Learning Rate l = 0.7
REMEMBER:
YOU HAVE TO SET YOUR INITIAL WEIGHTS AND BIASES
RANDOMLY; DON’T USE THE SET-UP FROM THE Lecture
EXAMPLE.