Question 1 [60 marks]

(1a1) Provide the mathematical formula of an error/loss function that is suitable for training neural networks in a classification problem. Explain the role of the various terms in this equation.

(5 marks)

(1a2) Provide the mathematical formula of an error/loss function that is suitable for training neural networks in a regression problem. Explain the role of the various terms in this equation.

(5 marks)

(1b) During the COVID- 19 outbreak, a local dealership launched online click and collect services, where customers can browse cars and order online, and then collect the car from the dealership office. The following figure shows data of customers’ visiting times and transactions just after a couple of days of online operation. The x-axis shows time in minutes that customers spent on the website. The y-axis shows the amount of money these customers spent in pounds (£K).

(1b1) How many features are used to describe customers’ behaviour in this case? Explain why.

(1 mark)

(1b2) Looking at these data, how many features does one have to  select in order to accurately predict a customer’s spending and why?

(1 mark)

(1b3) How can one measure in computational terms the similarity between customers of the same group and between customers of distinct groups?

(3 marks)

(1c) A fuzzy rule-based system is controlling the speed of a train. The system can smoothly slow and stop the train from any speed and distance from a station. The design of the system has been based on the following input and output variables and ranges:

(1c1) Based on the ranges and values shown above, explain what the universe of discourse of each input and output variable is. What could the fuzzy membership functions of the input and output variables be, if the system designers have used continuous triangular functions, or combination of continuous triangular and trapezoid functions, when defining fuzzy sets for each variable?

(10 marks)

(1c2) Give examples of two fuzzy rules that can be included in the rule base of the system to cover the scenario that the train is near a station.

(5 marks)

(1c3) Use the example you provided above to explain what rules can be triggered when train speed is 3km/h and distance to the station is 1.8m

(5 marks)

(1d) The weights of a neural network topology are encoded in a chromosome with 16 genes.

(1d1) Create your own example and explain how this approach can be applied on two different neural networks of the same topology, when the weights are real numbers in the interval (- 1, +1).

(10 marks)

(1d2) Use the two neural networks you have encoded above as a pair of parents, i.e. Parent- 1 and Parent-2. Apply uniform crossover with mask vector [0100110001001100], where 0 refers to inheriting a gene from Parent- 1 and 1 refers to inheriting a gene from Parent-2, to generate offspring, Child- 1 and Child-2. Show and explain the process you followed to create the children and the way these can be mapped back into neural networks.

(10 marks)

(1e) In an environmental AI application, a chemical sensor array measures a set of features for further processing by an intelligent system. One of the sensors appears to have a fault and  requires  maintenance urgently.  The  machine  learning  engineers believe  that  the sensing  environment  is  stable,  and  the  statistical  characteristics  will  not  change significantly during maintenance. Instead of taking the intelligent system off-line, they suggest using historic data to estimate the sensor’s input to the system and integrate it with other sensors data. From a set of 16 samples in a randomly selected time window they calculate a sample mean of 1.35 for the feature measured by this sensor, and from a longer set of 1,000,000 measurements from the operation of the sensor last month they estimate a mean of 1.4. To evaluate whether this plan is acceptable, they conduct a statistical hypothesis testing  at  0.05  significance  level, which produces the  acceptance  interval [1.237, 1.463]. Does the outcome of the hypothesis test support the suggestion to take the sensor off-line for maintenance and in the meantime feed the estimated mean value to the AI system? Explain your view. What is the percentage of confidence in that decision?

(5 marks)

Question 2 [20 marks]

(2a) Discuss three techniques for knowledge acquisition covered in the module. Explain how one could use each one of them to capture knowledge for modelling user behaviour in an interactive intelligent system.

(10 marks)

(2b) Explain the notion of a swarm. Use high-level description or pseudo-code to explain the overall operation of the particle swarm algorithm. Present each step of the algorithm and the operations that take place, explaining how one can use this method for training a neural network.

(10 marks)

Question 3 [20 marks]

(3a) Neuro-fuzzy systems have been proposed as a promising approach to build intelligent systems that would combine the advantages of both neural networks and fuzzy systems.

(3a1) What is a neuro-fuzzy system?

(2 marks)

(3a2) In the context of this module, discuss two alternative ways to combine neural networks and fuzzy systems to form a neuro-fuzzy system.

(6 marks)

(b) What is competitive learning? Derive a competitive learning algorithm presented in this  module,  describe  its  steps,  the  operators  used  and  explain  the  operation  of the algorithm.

(12 marks)