AI AI Engineer: Neural Networks and Deep Learning 3

Question 1

What distinguishes a generative adversarial network (GAN) from a variational autoencoder (VAE)?

Accepted Answer

GANs use an adversarial training loop between generator and discriminator; VAEs optimize a variational lower bound

Answer

GANs use supervised learning; VAEs use unsupervised learning

Answer

GANs encode data to latent space; VAEs generate from noise

Answer

GANs require labeled data; VAEs do not

Question 2

What is 'hyperparameter tuning' in deep learning?

Accepted Answer

Searching for optimal settings like learning rate, batch size, and architecture before/during training

Answer

Adjusting model weights during backpropagation

Answer

Removing unused neurons from a trained model

Answer

Scaling input features to unit variance

Question 3

In convolutional neural networks, what is the primary purpose of a pooling layer?

Accepted Answer

To reduce spatial dimensions and provide translation invariance

Answer

To add non-linearity

Answer

To increase the number of feature maps

Answer

To normalize activations

Question 4

What is 'early stopping' as a regularization technique in deep learning?

Accepted Answer

Halting training when validation loss stops improving to prevent overfitting

Answer

Stopping training when loss reaches zero

Answer

Reducing learning rate after a fixed number of epochs

Answer

Freezing certain layers during training

Question 5

Which technique is used to visualize which parts of an input image most influence a CNN's classification decision?

Accepted Answer

Grad-CAM (Gradient-weighted Class Activation Mapping)

Answer

PCA projection

Answer

t-SNE embedding

Answer

Kernel density estimation

Question 6

What is 'knowledge distillation' in the context of deep learning model compression?

Accepted Answer

Training a small 'student' model to mimic the outputs of a large 'teacher' model

Answer

Extracting features from raw data

Answer

Removing redundant neurons from a trained model

Answer

Converting weights to lower precision