Question 1

How does the Dueling DQN architecture enhance the learning of value functions?

Options :

A : By incorporating a second experience replay buffer for improved stability

B : By combining Q-values with advantage values to better separate value and advantage estimates

C : By adding a recurrent layer to capture temporal dependencies

D : By using a separate network to compute the discount factor for future rewards

Answer: B

Question 2

Which type of Monte Carlo method specifically evaluates the value of a policy based on the average returns of states visited under that policy?

Options :

A : On-Policy MC

B : Policy Gradient MC

C : Off-Policy MC

D : Model-Free MC

Answer: A

Question 3

In Keras, which class is used to define a custom training loop?

Options :

A : tf.keras.Model

B : tf.keras.callbacks.Callback

C : tf.keras.layers.Layer

D : tf.keras.optimizers.Optimizer

Answer: A

Question 4

Which GAN variant is specifically designed to improve stability and training efficiency?

Options :

A : Wasserstein GAN (WGAN)

B : CycleGAN

C : Conditional GAN (cGAN)

D : Deep Convolutional GAN (DCGAN)

Answer: A

Question 5

What is the main challenge associated with the variance of Policy Gradient estimates?

Options :

A : It can lead to slow convergence and unstable learning

B : It complicates the implementation of target networks

C : It requires large experience replay buffers

D : It makes the reward signals less informative

Answer: A

Free CAIE Mock Exam – Practice Online Confidently