Artificial Neural Networks (ANNs)

  • 19 different neural network architectures are applied.
  • The networks differed from each other in the number of hidden layers and in the number of neurons in a hidden layer.
  • Every node in a layer (input, hidden and output) in these feed-forward networks (multilayer preceptors) is fully connected (by adjustable weights) to every node in a subsequent layer.
  • Thus, the total input to a node of the hidden layer can be written as follows:

  • W are the adjustable connection weights.
  • Constant terms, so called biases are given by introducing an extra input to each unit.
  • Transfer functions are of the logsig type, thus a unit has a real-valued output: