Rprop

[dakofler]

Rprop is usually used for deep networks since backpropagation is slower for neurons farther away from the output. Pprop uses a variable learning rate to adapt to more steep or shallow areas in the gradient surface.