Does gradient descent in artificial neural networks apply the most changes closest to the input layer?

Yes. That's why backpropagation in deep networks is so slow.

Try watching cs231n at youtube

@Fast-Nop so, the problem is that it can't differentiate whether more changes in the next layer is needed versus the previous layers?

@BadFox actually, it's how chained partial differentials work. A very nice and free NN tutorial that makes the problem clear is http://neuralnetworksanddeeplearning.com/... and specifically chapter 5 (why deep NNs are hard to train), but that doesn't make much sense without having read the chapters before.

@Fast-Nop well, I guess I found my reading material for the next however long it takes. Thanks for the help. (^^)

@Fast-Nop.

Great link thanks!