THE UNIVERSITY OF SUSSEX CG019

BA & LLB FIRST YEAR EXAMINATION 2002

COGNITIVE MODELLING

Monday 9th September 2002 09.30 - 11.00

Candidates must attempt TWO questions
Each question is worth 25 marks.

1. 1. In the context of feed forward neural networks explain the terms: input layer, output layer, hidden unit, arc weight and node threshold. (10)
   2. A neural network that you have studied can compute the probabilities that a particular five letter sequence, such as groly is either an English word or not. Explain how the system is trained to produce such a result.
      (10)
   3. Supposing the word ``yacht'' is given a low probability of being an English word by the neural network in (b). What factors in the training regime are likely to have influenced this and how could the training be altered to improve the probability? (5)

2. In the context of production systems
   1. Explain the terms production rule, token, working memory, production memory, and conflict resolution. (10)
   2. Describe the cycle of operation of a production system. (10)
   3. Discuss the strengths and weaknesses of production systems as tools for cognitive modelling. (5)

3. 1. A program that you studied could simulate several children with differing errors in subtraction. Explain with examples how the program was able to vary its behaviour to model each child. (5)
   2. Using the Blocks World language understanding program that you have studied as an example, explain with examples the stages the program goes through in interpreting a command for the simulated robot to carry out. (10)
   3. Describe what a Braitenberg Vehicle is and explain, with an example, what they can be used to model. (10)