An ML engineer is training a deep network for image classification. The loss curve oscillates and the model fails to converge. The engineer suspects the learning rate is too high and the gradients are noisy. Which optimizer configuration is most appropriate to stabilize training while maintaining convergence speed?

Use SGD with a very high learning rate and no momentum

Use RMSProp with a very low learning rate and no momentum

Switch to Adam with a moderate learning rate and default beta values

Keep SGD but increase batch size to 1

What is the primary purpose of backpropagation in training a neural network?

To compute the gradient of the loss with respect to each network parameter

To propagate inputs forward through the network to generate predictions

To reduce overfitting by randomly dropping neurons during training

To initialize the weights of the network

A deep learning engineer is training a neural network with a single hidden layer. The loss function is mean squared error. During backpropagation, the gradient of the loss with respect to the hidden layer's weights is computed. The engineer notices that the gradients are extremely small, causing the hidden layer weights to update very slowly. Which of the following is the most likely cause?

Using too many neurons in the hidden layer

Using ReLU activation in the hidden layer

Using sigmoid activation in the hidden layer

A machine learning engineer is training a neural network with a large dataset. The training loss is decreasing smoothly, but the validation loss has been increasing for the last 10 epochs. The engineer wants to improve generalization. Which of the following actions should be taken?

Stop training immediately and use the current model

Use early stopping to revert to the best model

GIAC Machine Learning Engineer

Domain 3Objective 3

Neural Networks GMLE Practice Questions (Page 6)

Part of the Advanced Machine Learning and Neural Networks domain, which makes up ~30% of our current practice bank. GIAC (SANS) does not publish an official question count, but from its 180-minute exam (~70–120 total, ~21–36 in this domain), expect 7–12 from this objective — we provide 39 practice questions to prepare you well beyond it. (estimate)

39questions here

8free pages

9concepts

Questions 26–30

26application · medium
An ML engineer is training a deep network for image classification. The loss curve oscillates and the model fails to converge. The engineer suspects the learning rate is too high and the gradients are noisy. Which optimizer configuration is most appropriate to stabilize training while maintaining convergence speed?
Select an answer first
27foundation · easy
What is the primary purpose of backpropagation in training a neural network?
Select an answer first
28application · medium
A deep learning engineer is training a neural network with a single hidden layer. The loss function is mean squared error. During backpropagation, the gradient of the loss with respect to the hidden layer's weights is computed. The engineer notices that the gradients are extremely small, causing the hidden layer weights to update very slowly. Which of the following is the most likely cause?
Select an answer first
29foundation · easy
Which activation function is defined as f(x) = max(0, x) and is widely used in hidden layers to mitigate the vanishing gradient problem?
Select an answer first
30application · medium
A machine learning engineer is training a neural network with a large dataset. The training loss is decreasing smoothly, but the validation loss has been increasing for the last 10 epochs. The engineer wants to improve generalization. Which of the following actions should be taken?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

Next 5 questions →Back to GMLE

Free Basic Practice is a study aid with revealable answers — not a scored exam. Examers.io is independent and not affiliated with or endorsed by GIAC (SANS). “GMLE” is a trademark of its owner, used for identification only.

Neural Networks GMLE Practice Questions (Page 6)

Questions 26–30

What is the primary purpose of backpropagation in training a neural network?

Which activation function is defined as f(x) = max(0, x) and is widely used in hidden layers to mitigate the vanishing gradient problem?

A machine learning engineer is training a neural network with a large dataset. The training loss is decreasing smoothly, but the validation loss has been increasing for the last 10 epochs. The engineer wants to improve generalization. Which of the following actions should be taken?

Keep studying — Advanced Machine Learning and Neural Networks