Neural Nets

PART IV - bpNet Algorithm

bpNet Layout

Here is a simplified, watered-down version of the Neural Net algorithm. It demonstrates the basics in forward and backward propagation used for weight error correction.

(note: the source code accompanying this tutorial may be downloaded here.)

Backward Propagation

(note: vector[2] is the target value)

Forward Propagation

(note: vector[2] is the output)

Sigmoid

y = 1 / (1 + exp(-x))

Sigmoid Prime

y = x * (1 - x)

<Under Construction>

Next: EOF (aka no more)

Questions? Comments? Send them here. Or return to deXXed.