A generator is being trained on photographs of birds. Its score for a batch is the average "probability real" that the current discriminator gives the samples in it, and the generator is updated to push that average up. The discriminator is shown one sample at a time, as usual.
Two batches of eight samples are on the table — two things the generator could learn to produce.
Batch A — eight different birds. The discriminator scores them:
Batch B — the same bird, eight times. The discriminator scores every copy at .
Which way does this objective push the generator, and on what grounds?
Select all that apply.