Deep Learning Generative Models

Question 1

What are the two main components of a Generative Adversarial Network (GAN)?

Accepted Answer

Generator and discriminator trained in opposition to each other

Answer

Encoder and decoder

Answer

Policy network and value network

Answer

Inference network and sampling network

Question 2

What is mode collapse in GAN training?

Accepted Answer

The generator producing only a few distinct outputs rather than the full data distribution

Answer

The discriminator loss collapsing to zero

Answer

The GAN failing to converge during early training

Answer

The discriminator ignoring generated samples

Question 3

What is a Variational Autoencoder (VAE) and how does it differ from a standard autoencoder?

Accepted Answer

A VAE encodes inputs as probability distributions in latent space rather than fixed vectors, enabling principled sampling and generation

Answer

A VAE is identical to a standard autoencoder but uses convolutional layers

Answer

A VAE uses adversarial training instead of reconstruction loss

Answer

A VAE has no decoder, only an encoder

Question 4

What does the reparameterization trick in VAEs enable?

Accepted Answer

Backpropagation through the stochastic sampling step by expressing samples as a deterministic function of parameters and a noise variable

Answer

Using a different activation function in the encoder

Answer

Increasing the dimensionality of the latent space during training

Answer

Replacing the KL divergence term with cross-entropy

Question 5

What is a diffusion model in the context of generative deep learning?

Accepted Answer

A generative model that learns to reverse a gradual noising process to generate data from pure noise

Answer

A model that diffuses information across spatial dimensions using convolutions

Answer

A model that applies attention over a noise schedule to produce embeddings

Answer

A GAN variant using spectral normalization for stability

Question 6

What loss function does a standard GAN generator minimize?

Accepted Answer

The negative log-probability of the discriminator classifying generated samples as real

Answer

Reconstruction loss between generated and real images

Answer

Cross-entropy between generated distribution and target distribution

Answer

Mean squared error between latent codes and output images