Adversarial Autoencoders: The Encoder's Adversarial Loss
Intermediate
The problem, from first principles
In the AAE the encoder plays the generator's role in the adversarial game: it tries to make its codes look like prior samples to the discriminator. The loss is the same non-saturating form as a GAN generator loss.
From theory to code
Implement encoder_fooling_loss(d_codes), returning -mean(log D(z)) over the encoder's codes.
Constraints
- Returns a Python float.
Hints
Hint 1
Take the log of the discriminator outputs on codes, average, and negate.
Keywords
research-papersunsupervisedautoencodersadversarial
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.