1

   bp_innerloop.m Inner loop of the backpropagtion learning algorithm. One hidden layer. Uses tanh as the transfer function. Uses the following global variables for input and/or output: Inputs1 - input patterns Desired - desired output patterns LearnRate - learning rate parameter Momentum - momentum parameter DerivIncr - increment to the derivative of the transfer function (Fahlman s trick typical value 0.2) Weights1 - first weight layer (updated by this routine) Weights2 - second weight layer (updated by this routine) deltaW1 - initialize to 0 before first call deltaW2 - initialize to 0 before first call TSS - total sum-squared error (set by this routine) Recurrent state