Consider the graph in Figure 1 below, representing the training procedure of a GAN. Reference this figure for the next few questions.                                                                                                FIGURE 1

Reference Figure 1. You know that your GAN is trained when D(G(z)) is close to 1

Which of the following is common ground between GAN and VAE? (mark all that apply)

Given our objective is to maximize a quantity Z Z defined as:  Z = F(x,y) + G(x,y) + H(x,y) where F, G, H are functions with the following properties: F(x,y) has a closed form solution. G(x,y) has a closed form solution. H(x,y) is intractable. H(x,y)

Reference Figure 1. Early in the training, is the value of D(G(z)) close to 0 or close to 1?

Which of the following is common ground between GAN and VAE? (mark all that apply)

Given our objective is to maximize a quantity Z Z defined as:  Z = F(x,y) + G(x,y) + H(x,y) where F, G, H are functions with the following properties: F(x,y) has a closed form solution. G(x,y) has a closed form solution. H(x,y) is intractable. H(x,y)

Given random variables x, and y with distributions P(x) and P(y). Let D = KL(P(x) || P(y))       where KL is defined as the KL divergence (Kullback–Leibler Divergence) .   Which of the following statements is FALSE? (mark all that apply)

Reference the original problem. Now Suppose H(x,y) >= 0 instead of H(x,y)

Reference Figure 1. Early in the training, is the value of D(G(z)) close to 0 or close to 1?