Back to course
Next page
Interactive Quiz
Test your knowledge!
1
What is the main objective of a generative model in deep learning?
A
Categorize new input data
B
Learn the data distribution to generate new similar examples
C
Predict labels for input data
D
Reduce the dimensionality of data
2
In a GAN architecture, what is the role of the discriminator?
A
Generate realistic examples from a random vector
B
Classify whether an example is real or generated by the generator
C
Encode data into a regular latent space
D
Add noise to input data
3
What major problem can occur during GAN training, where the generator produces a low diversity of examples?
A
Overfitting
B
Mode collapse
C
Vanishing gradients
D
Overlearning
4
What is the main difference between a classic autoencoder and a variational autoencoder (VAE) in the latent representation?
A
The VAE encodes the input as a probability distribution (mean and variance), whereas the classic autoencoder encodes it as a single point
B
The classic autoencoder uses a more complex loss function than the VAE
C
The VAE does not include a decoder, unlike the classic autoencoder
D
The VAE cannot be used for data generation
5
What term is added to the VAE's loss function to ensure that the latent space follows a standard normal distribution?
A
Cross-entropy
B
Mean squared error
C
Kullback-Leibler divergence
D
Adversarial loss
6
What is the main advantage of Normalizing Flows compared to GANs and VAEs?
A
Their training is very stable and converges more easily
B
They produce sharper images than GANs
C
They do not require bijective functions
D
They have a latent space that is easily interpretable
7
In the diffusion process of Diffusion Models, what does the 'reverse process' represent?
A
Iterative addition of Gaussian noise to the image
B
Iterative removal of noise to reconstruct the original image
C
Transformation of the image into a low-dimensional latent vector
D
Classification of images based on their noise
8
What network architecture is generally used in Diffusion Models to predict the noise added to an image at each step?
A
Fully Connected (Dense) Network
B
Simple convolutional network
C
U-Net
D
Standard Transformer
9
What is the main disadvantage of Diffusion Models compared to other generative models like GANs?
A
Difficulty generating realistic images
B
Significant slowness in the generation process
C
Training difficulty and instability
D
Need for labels during training
10
In a Conditional GAN (cGAN), how are the attributes of the generated images controlled?
A
By modifying the size of the random input vector to the generator
B
By providing a label or conditional information as input to both the generator and the discriminator
C
By changing the loss function used during training
D
By using a deeper discriminator
Score: 0/10
Score: 0/10