I have a 3-2-1 ( 3 input nodes, 2 nodes/neurons in the hidden layer
and 1 neuron at the output layer), Ho can i compute the computation
requirement of this network ?
What i mean is that, how best can i know the computational order
( number of additions, number of multiplication ) of the network if i
am using sigmoid as the activation function.
GC.
This is what I would do:-
EasyNN-plus can trace function calls as it is training the network.
The individual functions can then be examined.
Steve
--
Neural Planner Software Ltd www.NPSL1.com
Neural network applications, help and support.
For an I-H-O MLP
Y = B2 + W2*tanh(B1+W1*X)
size(W1) = [H I]
size(B1) = [H 1]
size(W2) = [O H]
size(B2) = [O 1]
Just count the number of operations in the matrix equation.
Of course you will have to find out how many operations
are used to implement tanh. For MATLAB:
tanh uses FDLIBM, which was developed at SunSoft, a Sun
Microsystems, Inc. business, by Kwok C. Ng, and others.
For information about FDLIBM, see http://www.netlib.org
Hope this helps.
Greg