Name:              ___________________

Student No:      ___________________

Faculty:            ___________________

Signature:         ___________________

 

 

 

 

 

 

 

 

 

 

 

 

 

THE UNIVERSITY OF QUEENSLAND

FINAL EXAMINATION, SEMESTER II, NOVEMBER 1999

 

IC230

LABORATORY INTRODUCTION TO NEURAL NETWORKS

 

Time: Two (2) hours for working

Ten (10) minutes for perusal before examination begins

 

 

 

 

 

 

 

 

 

 

 

 

 

General Directions to Candidates

 

There are three sections to the paper, labeled “Neural Networks”, “Symbolic Models” and “Genetic Algorithms”. You should attempt to answer all three sections. The marks for each question are given in brackets after each question (Total marks 50).

 

The examination has been set to take the full two hours, so answer the questions that you know first, and only after answering them, go back to the questions that you are less certain about.

 

Answer all questions on the examination paper. If you need additional space use the back of the paper and clearly label each answer.

SECTION A: NEURAL NETWORKS (30 MARKS)

 

Question 1

 

a)      Draw a network of threshold logic units that implements the XOR function. Provide the values for all weights and biases. (2)

 

 

 

 

 

 

 

 

 

 

b)      What is the main difference between the the networks necessary for the XOR function as compared against the OR or AND functions? Why is this difference necessary? (1).

 

 

 

 

 

 

 

Question 2

 

Name and describe the three main learning paradigms. (2)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Question 3

 

a)      What are the defining characteristics of an IAC network? (1).

 

 

 

 

 

 

 

 

b)      Provide the activation equation of an IAC unit (1).

 

 

 

Question 4

 

The IAC Jets and Sharks network can be used to demonstrate a number of important properties of neural information processing. Define the following properties and describe how the Jets and Sharks network can be used to demonstrate this property.

 

a)      Content addressability (1):

 

 

 

 

 

 

 

 

b)      Robustness to noise (1):

 

 

 

 

 

 

 

 

Question 5

 

a)      What is the word superiority effect? (1)

 

 

 

 

 

 

 

 

b)      Describe briefly how the IAC model of letter perception explains the word superiority effect? (1)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Question 6

 

a)      Define the term “distributed representation”. (1)

 

 

 

 

 

 

b)      Provide two advantages of distributed representations. (1)

 

 

 

 

 

 

 

 

 

 

 

 

 

Question 7

 

Suppose we have the following associations that we wish to memorize using Hebbian learning:

 

Robin  (1 1 0 0 0) -> Flies (1 0)

Parrot (1 0 1 0 0) -> Flies (1 0)

Kangaroo (0 0 0 0 1) -> Jumps (0 1)

 

 

a)      Calculate the memory matrix to associate these patterns. (1)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

b)      If we cue the memory with the pattern for Emu (1 0 0 1 0) what output do we get and why? (1)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Question 8

 

a)      What is the difference between synchronous and asynchronous updating in a Hopfield network? (1)

 

 

 

 

 

 

b)     

Using the energy function:

 

Demonstrate that in the Hopfield network containing a single pattern, that pattern is a local minimum. (2).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Question 9

 

Use the perceptron convergence procedure to find a set of weights for a three unit network that implements the NOR function. (2)

 

NOR function

Input 1             Input 2             Target

1                      1                      0

1                      0                      0

0                      1                      0

0                      0                      1

 

 

W1

W2

Bias

I1

I2

Output

Target

0

-1

-0.5

1

1

 

0

 

 

 

1

0

 

0

 

 

 

0

1

 

0

 

 

 

0

0

 

1

 

 

 

1

1

 

0

 

 

 

1

0

 

0

 

 

 

0

1

 

0

 

 

 

0

0

 

1

 

Hint: The table shows how many steps it should take to find (and verify) the solution.

 

 

Question 10

 

a)      Why is the sigmoid activation function used in the backpropagation algorithm rather than the threshold logic function? (1)

 

 

 

 

 

 

 

b)      Give the equation for momentum. (1)

 

 

 

 

c)      What is the value of including a momentum term in the backpropagation training algorithm? (1).

 

 

 

 

 

 

 

 

 

 

Question 11

 

a)      Give the activation equation for the self organizing map. (1)

 

 

 

 

b)      What is a twisted map and how do they arise? (1).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Question 12

 

a)      What is the Stroop effect? (1)

 

 

 

 

 

 

 

 

 

 

 

 

 

b)      Describe how the Cohen, Dunbar and McClelland (1990) model explains the Stroop effect? (1)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Question 13

 

a)      In the field of human memory, what is the difference between a retrieval task and a matching task? (1)

 

 

 

 

 

 

b)      What is the difference between an episodic task and a semantic task? (1)

 

 

 

 

 

 

c)      Give examples of each of the different sorts of memory tasks in the following table: (1)

 

 

Matching Task

Retrieval Task

Episodic Task

 

 

Semantic Task

 

 

 

 

SECTION B: SYMBOLIC MODELS (10 MARKS)

 

Question 14

 

Soar is given a block stacking problem, starting with three blocks, A, B, and C. A is on top of B which is on the tabletop, and C is directly on the tabletop. Soar's goal is move the blocks one at a time to create a single stack of blocks, with A on top of B, and B on top of C. Using this problem as an example domain (3)

 

a)      Explain what an "impasse" is in Soar

 

 

 

 

 

 

 

 

 

 

b)      Describe a possible impasse situation