Sample Questions from Week 3 – Memory modeling

1. Network Design
   Let the items kangaroo, wallaby and echidna be represented by the patterns (1, -1, 1, 1) and (1, -1, -1, 1) and (-1, -1, -1, -1) respectively
   and the items grass and worms be represented by the patterns (1,1) and (1,-1) respectively.
1. Draw a neural network that could be used to model a simple recognition memory for the animal items.
2. Could the food items be stored in the same memory? Why or why not?
3. Draw a neural network that could be used to model a simple associative memory for the relations between the animals and the food they eat.
4. How many layers does the network have?
5. How many units are needed in each layer?
6. How many weights are in your network?
7. What learning rule would be appropriate?
8. What would be appropriate initial values for the weights?

2. Training
1. If the initial network was trained on the association between echidna and worms once, what value would the weights have?
2. If it was then trained another three times, what value would the weights have compared to after the first time?
3. If the initial network was trained on the list of associations, kangaroo and grass, wallaby and grass, and echidna and worms, and was tested on the pattern for kangaroo, what response would the network produce?

3. Properties of memory models
1. What is the difference between modelling recognition and association (or recall) in terms of the output of a memory model?
2. List two advantages and one disadvantage of distributed representations in cognitive modelling.
3. What aspects of Hebbian networks (or matrix models) are useful for modeling human memory?
4. Describe the Hebbian learning rule.
5. What outputs does a hebbian network give when trained on (i) orthogonal and (ii) non-orthogonal patterns? Explain your answer.