[LDA] basic knowledge notes - mainly AE and VAE

Bayes correlation

Likelihood function

Likelihood function （likelihood）：p(data| $\theta$)
Under the condition that the parameter value is determined , Probability of occurrence of current experimental data .

Posterior distribution

Posterior distribution ：p($\theta$|data)

Prior distribution

Prior distribution p($\theta$)
Like flipping a coin , The prior distribution should be a parabola , stay theta by 0.5 It's the biggest when I'm young .（ signify theta by 0.5 Is the most likely distribution of hidden variables ）

Bayes' formula ：

p($\theta$ | data)= [p(data| $\theta$)*p($\theta$)] / p(data)

AE Self encoder

Reference link csdn_ Self encoder (Auto-Encoder)

So understand ：
encoder: x->z
decoder:z->x’
loss=d(x,x’) d Represents the distance （ European style, etc ）
AE=encoder+decoder

VAE Variational self encoder

Auto-Encoding Variational Bayes（VAE）
Reference link csdn

“ and VAE Just one more move , That is to say, at the input of the encoder, we get latent space when , By doing Add noise , Enables the decoder to “ Accept ” A little unused latent space, Thus, more reasonable generated samples can be generated .”

Reference resources csdn_ Self encoder (Auto-Encoder)

Variational self encoder (Variational AutoEncoders,VAE)：
Given a hidden variable （ transcendental ） Distribution P(𝒛) , If you can learn the conditional probability distribution P(𝒙|𝒛), Then through the right Joint probability distribution P(𝒙, 𝒛) = P(𝒙|𝒛)P(𝒛) sampling , Generate different samples .

VAE Model for implicit variables 𝒛 The distribution of has explicit constraints , I hope the implicit variable 𝒛 Conform to the preset a priori distribution P(𝒛). In the design of loss function , In addition to the original reconstruction error term , Implicit variables are also added 𝒛 Distributed constraints