V(D,G)=Ex∼preal(x)[lnDθd(x)]A Good at recog. real img? =∫xpreal(x)lnDθd(x)dx=…=∫xpreal(x)lnDθd(x)+Ez∼N(0,1)[ln1−Dθd(Gθg(z))] Good at recog. fake img? + change of variables ∫zϕ(z)ln1−Dθd(Gθg(z))dz+∫xp^Gln1−Dθd(x)dx+p^G(x)ln1−Dθd(x)dx
where
Discrimminator D and parameters θd
Generator G and parameters θg
real distribution is x∼preal(x)
latent is simply N(0,1) but G transforms it into implicit distribution x∼p^G(x)