22nd LSI Design Contests-in Okinawa  Design Specification - 3-2

3-2. parameter calculation

Taking the differences value between the output and the supervisor, the neural network parameter will change simultaneously until the output are as same as supervisor signal.


Neural Network 3

Fig 3 : 4-layer model with parameter

Figure above shows the neural network model with each parameter in its respective layer. The definition of each parameter are as describe below:


  • ki is the input signal
  • wi j2 is the weight between i-th output of a input layer and j-th input of hideen layer 1.
  • bi2 is the i-th bias of hideen layer 1
  • zi2 is the i-th input of hideen layer 1
  • ai2 is the i-th output of hideen layer 1
  • wi j3 is the weight between i-th output of hideen layer 1 and j-th input of hideen layer 2.
  • bi3 is the i-th bias of hideen layer 2
  • zi3 is the i-th input of hideen layer 2
  • ai3 is the i-th output of hideen layer 2
  • wi j4 is the weight between i-th output of hideen layer 2 and j-th input of a output layer.
  • bi4 is the i-th bias of a output layer
  • zi4 is the i-th input of a output layer
  • ai4 is the i-th output of a output layer
<<Back                 Next>>