Wasserstein GAN Quiz

Answers and explanations appear as you take the quiz.

1. What distance metric does WGAN use to compare real and generated distributions?
   • Wasserstein (earth-mover's) distance
   • Kullback-Leibler divergence
   • Cosine similarity
   • Euclidean pixel distance
2. In WGAN, the network that was the discriminator is renamed to what, and why?
   • A 'judge' because it votes yes or no
   • A 'critic' because it outputs a continuous realness score rather than a probability
   • A 'sampler' because it generates noise
   • A 'filter' because it removes bad images
3. Why must the WGAN critic be constrained to be 1-Lipschitz?
   • To make it train faster on GPUs
   • To reduce the number of parameters
   • Because the Kantorovich-Rubinstein duality used to compute the Wasserstein distance requires bounded gradients
   • To force it to output integers
4. How did the original WGAN enforce the Lipschitz constraint?
   • By adding dropout
   • By using batch normalization only
   • By freezing the generator
   • By clipping the critic's weights to a small range
5. What problem does WGAN notably reduce compared to vanilla GANs?
   • Mode collapse and uninformative loss
   • Overfitting to the test set
   • Slow image loading
   • Excessive memory use
6. What improvement did WGAN-GP introduce over the original WGAN?
   • A larger generator
   • A gradient penalty replacing weight clipping
   • A new dataset
   • Reinforcement learning rewards
7. Why can Wasserstein loss provide a useful signal when real and generated distributions barely overlap?
   • It becomes a binary accuracy score
   • It no longer depends on the critic
   • The distance can vary smoothly as the distributions move, providing gradients before they overlap
   • It requires the generator to match every sample exactly
8. What does the WGAN-GP gradient penalty encourage the critic’s gradient norm to do on sampled interpolations?
   • Become exactly zero everywhere
   • Grow without bound
   • Match the generator’s parameter count
   • Stay near one