Structure Of Autoencoder In Deep Learning