Neural Networks and Genetic Algorithms

• Draw the structure and describe the operation of a processing element.
• Explain what is meant by a neural network being "fully feedforward connected".
• What (in general terms) are the tasks of each of the layers of a counter-propagation network?
• After sufficient training, what has a backpropogation network learned?

• Draw: inputs, weights, intermediate value, activation function, output.
  An intermediate value computed from the input values and weights. Output which is the activation function of the intermediate value.
• If each PE in a layer has an input from every PE in the previous layer, then the network is fully feedforward connected.
• Kohonen Layer: Each PE recognises a piece of the input space.
  Grossberg layer: Outputs will approximate the output vector for the piece of input space recognised by the ith Kohonen layer PE.
• A function from input vectors to output vectors.

• Consider four digit numbers, e.g., 3526, 8719, etc.
• New numbers can be formed by single point crossover, e.g., crossing over 3526 and 8719 at position 3 produces 3529 and 8716.
• Starting with the numbers 3526, 8719, 6222, and 1892, show one generation of a genetic algorithm that's aiming to generates the maximal number 8799, using the following random numbers for selection and crossover point ...
  • Parent selection 1 = 0.34
  • Parent selection 2 = 0.87
  • Crossover point for parents 1 and 2 = 3
  • Parent selection 3 = 0.66
  • Parent selection 4 = 0.19
  • Crossover point for parents 1 and 2 = 2
• If you are bored, do some more (no more marks though)

      Value  Fraction  Cumulative  Selection
-------------------------------------------------------------  
#1     8719      0.43        0.43    1 4
#2     6222      0.31        0.74    3
#3     3526      0.17        0.91    2
#4     1892      0.09        1.00
Total 20359

#1 & #3 @ 3 = 8716  3529
#2 & #1 @ 2 = 8722  6219