Cogs 2010 Stroop Network Sample Question

 

Cohen, Dunbar and McClelland’s Stroop model was originally developed to demonstrate that a learning model could account for the differences in responses for colour naming vs word reading.  The Stroop effect is also found in other tasks, such as reading numbers vs. counting them, although the effect is not as strong.

 

Design a network to illustrate a learning account of reading numbers vs. counting, based on the original design by CD&M. Your design should include

 

  1. The network architecture (i.e., two channels – one for counting and one for naming).
    1. Input units for the numerals 1-5 and counting 1-5
    2. Input units for directing attention
    3. hidden units as appropriate
    4. output units for the response 1-5
    5. connecting links between the layers of the network
    6. initial weights for the links, and whether they are fixed or modifiable

 

  1. how many units are in your network?
    1. input layer
    2. hidden layer
    3. output layer

 

  1. how many weights are in your network?
    1. Input to hidden
    2. Hidden to output
    3. Biases at the hidden and output layers

 

  1. training and test sets:
    1. what patterns would you train the network on?
    2. What patterns would you test the network on?

 

  1. In the original CDM model, word reading was much faster than colour naming.  In reading numbers, the difference is not so large.  How do you model the smaller difference in your design compared to CDM’s model?