UFCFY3-15-3 JAN 15 EXAM PAPER
Faculty of Environment & Technology Academic Year: Examination Period: Module Leader: Module Code: Module Title: Work Item Code: Duration:
14/15 January
L Bull UFCFY3-15-3 ADVANCES IN ARTIFICIAL INTELLIGENCE EX1 2 Hours
Standard materials required for this examination: Examination Answer Booklet
Yes
Multiple Choice Answer Sheet
No
Graph Paper
Type of paper e.g. G3, G14
G3
Number of sheets per student
0
Additional materials required for this examination: Details of additional material supplied by UWE:
To be collected with Answer Booklet (please delete as appropriate)
N/A
a
Details of approved material supplied by Student:
To be collected with Answer Booklet (please delete as appropriate)
N/A
University approved Calculator
Yes
Candidates permitted to keep Examination Question Paper
Yes
Candidates are NOT permitted to turn the page over until the exam starts
UFCFY3-15-3
Page 1 of 5
Instructions to Candidates: Candidates must answer any FOUR questions.
1. The genetic regulatory processes within living cells can be abstracted into networks of nodes which use Boolean update functions. These can be used both to model aspects of natural systems and as a dynamic representation scheme within artificial systems. As a simple example consider the following network:
1 AND
2 OR
3 OR
a) Assuming nodes 1 and 2 are set to logical ‘1’ and node 3 is set to logical ‘0’ upon initialization, give the series of states each node will go through for the first two update cycles. (6 marks)
b) What do you notice about the state the network gets into at this time? (2 marks) c) Together with a rough sketch, show how these types of network can be used for computation, in particular for a problem with two inputs and one output. (4 marks)
UFCFY3-15-3
Page 2 of 5
d) Describe how evolutionary computation can be used to design these networks to perform a particular computation. (13 marks)
2. Advances in cell biology mean that it is now possible, in principle, to compute with living substrates, such as neurons. a) Describe each of these two pieces of equipment and advantages and disadvantages for neuronal computing for each:
(10 marks)
b) Describe a simple training protocol for neuronal networks based on the Stimulus Regulation Principle put forward by behaviourists. (15 marks)
3. The ideas behind swarm intelligence have been used in robotics, in what is often termed “collective robotics”. a) Describe the control scheme typically used for each robot in the swarm and give an example for a carpet vacuuming robot. (15 marks) b) What is evolutionary robotics and what potential benefits and problems are associated with it, especially for collective robotics? (7 marks) c) Give three examples of possible applications of collective robotics in the future. (3 marks)
UFCFY3-15-3
Page 3 of 5
4. Reinforcement Learning has proven useful in many areas. a) Using the standard Q-learning equation below, briefly explain what reinforcement learning entails.
Q(st ,at) ← Q(st ,at) + β [ rt+1 + γ max Q(st+1 ,at+1 ) – Q(st,at) ] (6 marks)
b) Starting in the middle state no.3 of the simple maze below and choosing the ‘right’ action on every step, show how Q-learning would adjust the value estimations of the possible state-action pairs during two such learning episodes. Assume a learning rate of 0.1, a discount rate of 0.8 and a reward at the goal state of 100. (15 marks)
c) Give two examples of tasks suitable to use reinforcement learning and two that are not, briefly stating why in each case. (4 marks)
5. Consider the following population of Genetic Programming solutions: UFCFY3-15-3
Page 4 of 5
1.
Fitness = 169
2.
Fitness = 576
3.
Fitness = 64
4.
Fitness = 169
a) Show a likely full population of parents after selection using the roulette-wheel scheme, indicating why. (10 marks) b) Show how this parent population might look after crossover using the traditional one-point scheme. (5 marks) c) Show how this population might be further changed under mutation. (5 marks) d) Discuss how the application of these three processes results in effective search. (5 marks)
END OF QUESTION PAPER
UFCFY3-15-3
Page 5 of 5
14/15 January
L Bull UFCFY3-15-3 ADVANCES IN ARTIFICIAL INTELLIGENCE EX1 2 Hours
Standard materials required for this examination: Examination Answer Booklet
Yes
Multiple Choice Answer Sheet
No
Graph Paper
Type of paper e.g. G3, G14
G3
Number of sheets per student
0
Additional materials required for this examination: Details of additional material supplied by UWE:
To be collected with Answer Booklet (please delete as appropriate)
N/A
a
Details of approved material supplied by Student:
To be collected with Answer Booklet (please delete as appropriate)
N/A
University approved Calculator
Yes
Candidates permitted to keep Examination Question Paper
Yes
Candidates are NOT permitted to turn the page over until the exam starts
UFCFY3-15-3
Page 1 of 5
Instructions to Candidates: Candidates must answer any FOUR questions.
1. The genetic regulatory processes within living cells can be abstracted into networks of nodes which use Boolean update functions. These can be used both to model aspects of natural systems and as a dynamic representation scheme within artificial systems. As a simple example consider the following network:
1 AND
2 OR
3 OR
a) Assuming nodes 1 and 2 are set to logical ‘1’ and node 3 is set to logical ‘0’ upon initialization, give the series of states each node will go through for the first two update cycles. (6 marks)
b) What do you notice about the state the network gets into at this time? (2 marks) c) Together with a rough sketch, show how these types of network can be used for computation, in particular for a problem with two inputs and one output. (4 marks)
UFCFY3-15-3
Page 2 of 5
d) Describe how evolutionary computation can be used to design these networks to perform a particular computation. (13 marks)
2. Advances in cell biology mean that it is now possible, in principle, to compute with living substrates, such as neurons. a) Describe each of these two pieces of equipment and advantages and disadvantages for neuronal computing for each:
(10 marks)
b) Describe a simple training protocol for neuronal networks based on the Stimulus Regulation Principle put forward by behaviourists. (15 marks)
3. The ideas behind swarm intelligence have been used in robotics, in what is often termed “collective robotics”. a) Describe the control scheme typically used for each robot in the swarm and give an example for a carpet vacuuming robot. (15 marks) b) What is evolutionary robotics and what potential benefits and problems are associated with it, especially for collective robotics? (7 marks) c) Give three examples of possible applications of collective robotics in the future. (3 marks)
UFCFY3-15-3
Page 3 of 5
4. Reinforcement Learning has proven useful in many areas. a) Using the standard Q-learning equation below, briefly explain what reinforcement learning entails.
Q(st ,at) ← Q(st ,at) + β [ rt+1 + γ max Q(st+1 ,at+1 ) – Q(st,at) ] (6 marks)
b) Starting in the middle state no.3 of the simple maze below and choosing the ‘right’ action on every step, show how Q-learning would adjust the value estimations of the possible state-action pairs during two such learning episodes. Assume a learning rate of 0.1, a discount rate of 0.8 and a reward at the goal state of 100. (15 marks)
c) Give two examples of tasks suitable to use reinforcement learning and two that are not, briefly stating why in each case. (4 marks)
5. Consider the following population of Genetic Programming solutions: UFCFY3-15-3
Page 4 of 5
1.
Fitness = 169
2.
Fitness = 576
3.
Fitness = 64
4.
Fitness = 169
a) Show a likely full population of parents after selection using the roulette-wheel scheme, indicating why. (10 marks) b) Show how this parent population might look after crossover using the traditional one-point scheme. (5 marks) c) Show how this population might be further changed under mutation. (5 marks) d) Discuss how the application of these three processes results in effective search. (5 marks)
END OF QUESTION PAPER
UFCFY3-15-3
Page 5 of 5
