CS534 ARTIFICIAL INTELLIGENCE. SPRING 98
PRACTICE PROBLEMS FOR FINAL EXAM
PROF. CAROLINA RUIZ
Department of Computer Science
Worcester Polytechnic Institute
 Reasoning under Uncertainty
 Russell and Norvig: 14.1, 14.3, 14.4,14.5,14.8,14.9,15.1,15.2,15.3.
 Additional Problems
 Machine Learning  Decision Trees
 Russell and Norvig: 18.3, 18.6, 18.11
 Dean, Allen, and Aloimonos:
Construct a minimal decision tree for the following set of
training examples that helps a robot janitor predicting whether or not
an office contains a recycling bin.
 STATUS  FLOOR  DEPT.  OFFICE SIZE
 RECYCLING BIN?

1.  faculty  three  ee  large  no

2.  staff  three  ee  small  no

3.  faculty  four  cs  medium  yes

4.  student  four  ee  large  yes

5.  staff  five  cs  medium  no

6.  faculty  five  cs  large  yes

7.  student  three  ee  small  yes

8.  staff  four  cs  medium  no

 Machine Learning  Neural Nets
 Russell and Norvig: 19.3, 19.4, 19.6
 Machine Learning  Genetic Algorithms
 Winston
 Suppose that you are testing various tent designs and tent materials
for comfort in winter. The temperature outside the tent is 40 degrees.
The temperature inside a tent of one designmaterial conmbination,
warmed by your body heat and a small stove is 5^{o}C. Ohter
tents produce temperatures of 10^{o}C and 15^{o}C.
 Compute the chances of survival of each combination using the
standard fitness method with inside temperature as the measure of quality
(the warmer the better).
 Now suppose that you have lost your Celsius thermometer, reducing
you to measuring temperature with a Fahrenheit thermometer. Recalculate
the chances of survival of each combination with the 41^{o}F,
50^{o}F, and 59^{o}F temperatures.
 Comment why your results argue against the use of the standard method.
 Suppose that you are using the rank method to select candidates.
Recall that the probability of picking the firstranked candidate is p.
If the first candidate is not pricked, then the probability of picking
the secondranked candidate is p (that is the second candidate's chance
of survival is p*(1p)), and so on, until only one candidate is left,
in which case it is selected.
 Suppose that you have five candidates. You decide to select
one using the rankspace method with p=0.25. Compute the probabilities
for each of the five as a function of rank.
 What is peculiar
about the probabilities that you computed in the previous part?
 Can you avoid the pecularity you observed in part 2 by restricting the
probability to a particular range?
 Dean, Allen, and Aloimonos: Discuss how you would represent and
solve the traveling salesperson problem using genetic algorithms.
 Machine Vision
 Russell and Norvig: 24.2, 24.3, 24.4, 24.5, 24.7