Can discriminative models be sampled from if we extract vectors within the hypothesis?

Question

Generative models are useful because we can sample vectors from them.

Discriminative models supposedely don't have this feature. But what's stopping us from sampling vectors that are "within the hypothesis"? Imagine we draw a circle around the cat vectors. And we can then randomly sample points within this circle to "generate" sample vectors of a cat.

So, by this technique: discriminative models can "generate" samples. I don't know if this technique is valid though

Answer 1

Do you know what we call the outputs of a model that provides us with the encoded vector and the radius of the circle (ellipsoid, technically) we can draw around the vectors and still get correctly generated samples?

We call them the 'mean' and 'standard deviation' (respectively).