CS5720 - GAN Training Process

Training Steps

Initialize Networks

Initialize both Generator and Discriminator with random weights

Train Discriminator

Update D to maximize log(D(x)) + log(1 - D(G(z)))

Train Generator

Update G to minimize log(1 - D(G(z))) or maximize log(D(G(z)))

Repeat & Monitor

Alternate training until convergence or satisfactory results

Generator Loss

Discriminator Loss

Epochs Completed

min_G max_D V(D,G) = E_{x~p_data(x)}[log D(x)] + E_{z~p_z(z)}[log(1 - D(G(z)))]

🎨

Generator's Strategy

Minimize the objective by making D(G(z)) as close to 1 as possible, fooling the discriminator into thinking fake samples are real.

🔍

Discriminator's Strategy

Maximize the objective by correctly classifying real samples (D(x) → 1) and fake samples (D(G(z)) → 0).

Nash Equilibrium:

At convergence, the Generator produces samples indistinguishable from real data, and the Discriminator outputs 0.5 for all samples.