Each of the following questions is worth 5 marks. Please write approximately 10-15 lines per question. Use diagrams where appropriate.
(a) Describe how the learning rate and momentum parameters of the back propagation algorithm affect an artificial neural network's traversal of an error surface. Use a diagram to support your explanation.
(b) The "Red Queen" is one concept associated with co-evolutionary system dynamics. Describe the "Red Queen" scenario in relation to a predator-prey system, and describe how a co-evolutionary system can be used to learn to play a zero-sum game (such as tic-tac-toe).
(c) The figure below shows three different training runs (labelled A, B & C) when applying backpropagation to your ANN. Comment briefly on the likely quality of the three resulting ANN's based on these errors.
(d) Assume that you have an evolutionary algorithm using roulette-wheel selection, crossover and no mutation. Explain why in general as the evolution proceeds the selection pressure of the population alters over time. What is the selection pressure once the population has converged?