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 pit a generator against a discriminator in an adversarial game, while VAEs learn a probabilistic latent space by maximizing an evidence lower bound (ELBO).

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

Hyperparameter tuning involves systematically searching over configuration choices (learning rate, depth, dropout rate) that are set before training, not learned.

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

Pooling (e.g., max pooling) downsamples feature maps, reducing computation and providing some invariance to small spatial translations.

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

Early stopping monitors validation loss and stops training when it starts increasing, saving the model at the point of best generalization.

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

Grad-CAM uses gradients flowing into the final convolutional layer to produce a heatmap highlighting the regions most important to the prediction.

AI - Engineer Practice Test

AI - Engineer Practice Test

AI AI Engineer: Neural Networks and Deep Learning 3