- : Data we already have
- : Data we want to generate
- : parameter for Generator
- : parameter for D
- : Generator
- : Discriminator
target is training to minimize
For player , want V bigger by
- making more accurate estimate on real date x as
- discriminate the fake data by equals to
- more accurate the is ,the larger value of can be
For player , want V smaller by
- enlarge for
- better fake of , , and less value of
Algorithm
for number of iterations do
for steps do
- Sample noise samples from noise prior
- Sample examples from data generating distribution
- Update the discriminator by ascending its stochastic gradient :
end for
- Sample noise samples from noise prior
- Update the generator by descending its stochastic gradient :
end for
ew
- q
- q