I'm training a NN with millions of iterations (100 million +). I've implemented my own NN and since it took so long to do the training, I decided to look into multithreading my NN (eg: evaluate the activation function of different neurons on the same layer simultaneously). I found that creating, starting and stopping threads often take longer than the function it executes (eg: activation function). So currently my multithreaded NN takes longer than my single-threaded one.
My question: does FANN make use of multiple threads on a single NN (I'm not taking about training multiple NN in parallel)? If not, why not (is it something similar to my problem above)?