【AREA-introduction】Generative models
Introduction of common generative models. (building)
general
Most of the datas such as audios,texts,images in our daily life have a lot of redundant informations.
VAEs
AE
Its biggest difference with normal networks is that there are much fewer nodes at its hidden layer,
which is like a bottleneck,that forced the network to learn about the specific representation of the input.
The encoder would be trained based on the difference between the reconstructed output and input.
But,due to the overfitting of the encoder,we could not possibly use the decoder as a generator.
VAE
The main task of VAE’s encoder is to learn how to represent 2 fractors mean and variance of the origin distribution with
a new distribution at hidden layer level,that is,P(X|Z)
By