The form of a single layer feed forward neural network lends itself to finding the gradient. This is useful when the network is used for surrogate optimization or other algorithms that use gradients. Requires creating a file by modifying a NN toolbox file.
I am working with 2013 and I modified 'tansig.m' with its derivative as d= 1-a.*a
but there is an error in 'nnModuleInfo'; saying '' undefined function or variable 'dtansig_0.name' ''
i want to download it
This needs a change to work with 2010 (when you change the size of the output, the weights change differently then before), I've done it and will get it in here soon. Again, Please let me know if this has helped you.
Let me know if a Hessian would be helpful. I'm sure it could be done, but I'm wondering if there's interest.
Download apps, toolboxes, and other File Exchange content using Add-On Explorer in MATLAB.