GAN: The Discriminator Loss
Intermediate
The problem, from first principles
A GAN (Goodfellow et al., 2014) trains a discriminator to tell real samples from generated ones. The discriminator's loss is the binary cross-entropy: score real data high and generated data low.
From theory to code
Implement gan_discriminator_loss(d_real, d_fake), the sum of the two cross-entropy means.
Constraints
- Outputs are probabilities strictly between 0 and 1.
Hints
Hint 1
Average -log D over real samples and -log(1 - D) over fake ones, then add.
Keywords
research-papersunsupervisedgangenerative-models
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.