In contrast to the same ganglion, small depth greater than the depth difference between the neural network neural network effect
which is why?
Modularization (modular)
As we write a program, the main function does not include all of the code, write some sub-functions so that it can be repeated multiple calls to the same function, deep learning, too.
For example, to train the example above, the data may be little long-haired boys, training effect is not very good, but it can be solved by a modular
structure of the human voice, Phoneme (phonetic), the words may be pronounced differently in the statement on there are different models, such as Tri-phone, a phonetic split into several state and can
pronounce all the vowels only with the human tongue back and forth, up and down and lips three factors
in depth study before a phoneme is a model, DNN is with a network, the neural network can be seen that the lower first read the pronunciation of the way, which played a modular role
if our NN simply linear not separate properly the figure below four points, but added a nonlinear as if the coordinates on the break, the shear bars as NN, to cut the traces is data, data tells us to cut the traces of those paintings you can get complex patterns, so you can use the good results of a small amount of data, NN also possible as much as possible to use data through deeper layers.
As another example