School of Computing. Dublin City University.
Online coding site: Ancient Brain
Temperature 22 degrees, pressure 1000 mb. Rain.
Temperature 15 degrees, pressure 990 mb. No rain.
Then the question is:
Temperature 17 degrees, pressure 1010 mb. Will there be rain?
We predict by Hamming distance to the nearest exemplar, (a) with pressure in millibars, (b) with pressure in bars. What does it predict?
Have a relationship with n partners of fitness > f
Marry the next partner you find of fitness > g
where fitness is ranked by the individual out of 10.
Which of the following values for n,f,g (respectively) makes the most sense?
i 1 2 3 4
f(i) 0 5 4 5If we use a "Soft max" probability distribution with a "temperature" parameter T to choose who reproduces, then as T > infinity the probability of each individual reproducing goes to:
0 1/2 0 1/2
0 1 0 1
0 5/14 4/14 5/14
0 0.9 0.1 0.9
0 1 0 0
1/4 1/4 1/4 1/4
Have a relationship with n partners of fitness > f
Marry the next partner you find of fitness > g
where fitness is ranked by the individual out of 10.
Which of the following values for n,f,g (respectively) makes the most sense?
a x + b y = cDefine a perceptron that receives x and y as input and fires if and only if the point is on this side of the line:
a x + b y ≤ c
The question is:
Opponent height  185 cm. Opponent weight  94 kg. Will I win?
We predict by Hamming distance to the nearest exemplar, (a) with height in cm, (b) with height in m. What does it predict?
An example showing that this is this a bad measure of the error is:
Start with 100 random solutions.
Repeat:
Test all 100 for fitness.
Pick the 10 best.
Make 10 copies of each to construct the next generation.
Loop again.
The problem with this is:
Question 1  [Total marks: 20] 
1(a) 
Chatbots talk about the physical world but they have no experience of it.
They might talk about "chairs" but they have never touched a chair or sat in a chair.
Discuss whether this matters or not.
Maximum length: 2 sentences 
[5 marks] 
1(b)  AI people love competitions. Why? Maximum length: 2 sentences 
[5 marks] 
1(c)  Machines beat human champions in chess but not in tennis.
Suggest an explanation why,
focusing entirely on the inputs to the system.
Maximum length: 3 sentences 
[5 marks] 
1(d)  A typical machine learning algorithm converges to a solution in a fairly short time.
Discuss whether human learning converges in a similar way or not.
Maximum length: 2 sentences 
[5 marks] 
Question 2  [Total marks: 20] 
2(a)  Mathematicians have longstanding tools to maximise functions
and calculate function values.
What is different about the "functions" AI programs try to maximise and calculate values for? Maximum length: 3 sentences 
[5 marks] 
2(b) 
Is this function y = f(x) easy or hard to maximise from exemplars? f(1.0372) = 80. f(5.6063) = 100. f(x) = 0 for all other x. Maximum length: 3 sentences 
[5 marks] 
2(c) 
Consider a heuristic search as follows:
Initialise with 20 random solutions. Repeat for many iterations: { Test the 20 solutions. Pick the best n. Make small changes in them to generate 20 new solutions. }If n=19 what will this look like? Maximum length: 2 sentences 
[5 marks] 
2(d) 
Discuss trying to maximise the following function from exemplars.
Explain what would happen during learning.
Input x is a number. It does not include the clock time. The function is defined as follows: if ( x > 10 ) f(x) = 1 if clock time is PM. f(x) = 1 if clock time is AM. else if ( x < 10 ) f(x) = 1 if clock time is PM. f(x) = 1 if clock time is AM. else f(x) = 0 Maximum length: 3 sentences 
[5 marks] 
Question 3  [Total marks: 20] 
3(a) 
In the Npuzzle problem, we use a heuristic (distance to goal estimate) as follows:
h(n) = number of pieces out of place.
Is this heuristic optimistic or pessimistic? And how is that useful? Maximum length: 2 sentences 
[5 marks] 
3(b) 
What would Bestfirst search with a constant heuristic evaluation function look like?
(A heuristic evaluation function that outputted a constant number c.)
Maximum length: 2 sentences 
[5 marks] 
3(c) 
When a problem is represented as a State Space Search problem,
a data structure
is constructed to represent
the entire "search tree". True or false?
Maximum length: 2 sentences 
[5 marks] 
3(d) 
In the Npuzzle problem, we use a heuristic (distance to goal estimate) as follows:
h(n) = number of pieces out of place.
If h(x) = 1 then are we close to the goal? Maximum length: 4 sentences 
[5 marks] 
Question 4  [Total marks: 20] 
4(a) 
Consider the character recognition problem for handwriting.
If a character is isolated from other characters, there may be a problem with missing inputs when we try to recognise it. Explain. Maximum length: 3 sentences 
[5 marks] 
4(b) 
A mobile robot has
Input state = (angle of direction of nearest wall, velocity value).
Output behaviour = one of (turn left, turn right, stop, keep going).
The Input state is clearly missing some information. What are the consequences of this missing information? Maximum length: 3 sentences 
[5 marks] 
4(c) 
In neural networks, why is it not a constraint to say that all networks
are fully connected?
Maximum length: 1 sentence 
[5 marks] 
4(d) 
The following is the perceptron learning rule.
We have multidimensional input,
where each component of the input I_{i} is binary 0 or 1.
It produces output y = 0 or 1.
We show it the "correct" output O = 0 or 1.
We modify the weights and the threshold by the
rules:
w_{i} := w_{i} + C (Oy) I_{i}
t := t  C (Oy)
where C is some positive constant.

[5 marks] 
Question 5  [Total marks: 20] 
5(a) 
In a genetic algorithm (GA),
let p_{c} be the probability of crossover.
What would happen if p_{c} = 1? Maximum length: 2 sentences 
[5 marks] 
5(b) 
In a genetic algorithm (GA),
let p_{m} be the probability that any bit in the
chromosome mutates
at the point where it is inherited from the parent by the child.
Let the chromosome be binary, so that mutation means a bit flip.
What would happen if p_{m} = 0 (but you still had crossover)? Maximum length: 4 sentences 
[5 marks] 
5(c) 
In a genetic algorithm (GA),
let each individual i have fitness f(i),
and let its probability of reproducing, p(i), be:
where T is the "temperature".

[5 marks] 
5(d) 
In a genetic algorithm:
What would happen if rather than picking the fittest, you define a minimum fitness that individuals have to reach
in order to reproduce?
Maximum length: 1 sentence 
[5 marks] 
Class:_____________________
Seat No:_____________________
