Adversarial Autoencoders: Sampling the Prior

The problem, from first principles

The adversarial autoencoder matches the distribution of its latent codes to a prior it can sample from. The prior here is a standard normal, and drawing samples from it is the first step of generating new data.

From theory to code

Implement prior_sample(rng, n, d), returning n standard normal samples of dimension d.

Constraints

• Use the supplied rng for reproducibility.

Hints