Making thresholds part of the weight matrix
Submitted by mancer on Mon, 03/03/2008 - 01:23
Hello, enjoying the book "Instroduction to Neural Networks with Java" so far. I have a question about feedforward backpropagation networks. I read somewhere that the threshold values can be made a part of the weight matrix. How would I go about modifying JOONE to do it this way, it seems like an more easier appraoch.
Mancer
Thats not a bad way to do it!
I am working on a neural network currently that is structured in that way. It is nice, because the entire memory of the neural network is inside of the weight matrix. Its not too hard to do, you just add another row or column(depending on how you map your matrix) to the matrix that represents the thresholds.
Jeff
Won't that mess up the input though?
But won't that mess up the input? Say its normally a three neuron input layer. Doesn't it now become a four? What do you put in for the forth neuron? I don't understand.