Denoise Autoencoder Simple Comprehension

since the encoder expresses the original data by learning the hidden features, what is Denoise autoencoder?

About Autoencoder Reference: http://blog.csdn.net/on2way/article/details/5009508

http://blog.csdn.net/on2way/article/details/50390595

Reference: http://www.cnblogs.com/tornadomeet/p/3261247.html

In order to enhance the robustness, the random noise can be added to the input layer, and the original data is reconstructed with the corrupted data when the unsupervised method is used to train the network in a layered way. The method of introducing random noise is

make some inputs randomly set to 0, which very similar to Droupout. The difference is thatdropout is not involved in the process of layering pre-training weights, but is only introduced by the fine-tuning in the back; Denoise autoencoder is Each layer is introduced as an input layer during the pre-training process and is not involved in fine-tuning.

Denoise Autoencoder Simple Comprehension