Random Noise in GANs

Question

What does the "random input" from noise (random noise) mean in GANs?

Let's say I want to synthesize data such as name, age, income etc.. Does my generator know the original data sets (and thus the distributions of the data)?

How can I think of "random noise" in this scenario?

Effectively what it says is that GANs generate new data which follow some distribution (learned from old data), so novelty requires some quite simple random input. Then it goes on to say that not any random input will do, it has to match the output dimensions — Nikos M., Oct 23 '21 at 09:31
Welcome to SO; please do not clutter your posts with irrelevant and unhelpful introductions, sorrys. thanks etc (edited out) — desertnaut, Oct 23 '21 at 23:35

score 0 · Accepted Answer · answered Oct 23 '21 at 21:42

In Generative Adversarial Networks (GANs), the random noise is weighted by the learned feature maps to generate novel, synthetic data examples.

If there was no random input to the generator network, it would create the same synthetic data example over and over. The noise acts as a random seed.

Random Noise in GANs

1 Answers1