First, I would like to thank you for the excellent article. It was a pleasure to read it. I gave it a five.
I have a comment regarding the problem with page faults in allocating the memory for Weights.
Indeed, to keep all network Weights in one vector or array or other container asks for a huge bulk of the memory.
The solution, in my opinion, is to use std::map<NeuronFromPrevLayer*, Weghts> or stdext::hash_map for each neuron inputs connection. This map or pointer to it should be a member of the Neupon class. You can find this design in An' Oop articles on BrainNet on the CodeProject.
Again, thank you for the excellent work.
Last Visit: 31-Dec-99 19:00 Last Update: 29-Mar-17 1:46