The best reading experience please go to the original address: variational self-encoder (variational autoencoder, VAE) Popular tutorial-Deng Fanxin
1. Mysterious variables and data sets
Now there is a DataSet dx (dataset, also called datapoints), and each data is also called a data point.
X is an actual collection of samples, and we assume that this sample is manipulated by some mystical force, but we don't know what these mystical forces are. Then we assume this mysterious force has N, the name is Power1,power2,..., Powern Bar, their size is z1,z2,..., Zn, called the mysterious variable expressed as a vector is
Z also has a name called the mystery combination.
Word: Mystical variables represent mystical combinations of mystical forces.
In a serious word, the implicit variable (latent variable) represents the combined relationship of the hidden factor (latent factor).
Here we clarify the membership space, assuming that the data set DX is M points, the M point should also be subordinate to a space, such as a one-dimensional situation, if each point is a real number, then his membership space is a real set, so we define a DX each point belongs to the space called XS, we mentioned later, You will no longer feel unfamiliar.
Mysterious variables Z can be sure that they also have a attribution space called ZS.
Now we are going to formalize the mysterious relationship between X and Z, which is the mystical power we've been talking about, and intuitively we're pretty clear that assuming our data set is completely manipulated by these n mysterious variables, Then for each point in X should have a mysterious combination of n mysterious variables zj to the mysterious decision.
Read more about the original: variational self-encoder (variational autoencoder, VAE) Popular tutorials-Deng Fanxin
Variational self-encoder (variational autoencoder, VAE) Popular tutorials