Too much limitation of Gaussian model. The images are too blurry. So any general model?

 

But if P_G(x;θ) is a neural network, it's impossible to calculate the likelihood. ????

 

 

 

 

 We don't know the formulation of PG and Pdata, so how to calculate the divergence???? ---->>> discriminator!!!

 

 

 

 

 

 

 

 

V(G,D) = maximum the output if data come from Pdata, and maximum the output of data from PG. --- >>> this process is identical to a binary classifier

 

 

 

 

 

 

 

 

 

 

 

 

不是minima 和 saddle point

而是maxima

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

even if L(G) is not differentiable (a Max operation), the derivative is computable.

 

 

 

 

 

 

Tip:

Train D as much as possible

Train G only to a moderate level

 

 

 

 

 

 

  

 

 

 

 

 

 

the results are actually similar......

  

 

 

 

 

 

 

 

 

green point is true data, blue point is from genrator

 https://www.youtube.com/watch?v=ebMei6bYeWw

 

 

 

 

Some one would argue that discriminator shouldn't initialize with the last discriminator, but the operation also sounds reasonable on another perspective.

Some one in a paper shows that the performance increases if the samples come from current and past generators, although it may not sound reasonable.