A deep CNN is training slowly and the activations in the hidden layers have widely varying scales across batches. The team suspects internal covariate shift. Which technique should they apply?

Increase the learning rate to speed up convergence.

Use a smaller batch size to reduce variance.

Add batch normalization layers after the convolutional layers.

Use dropout to reduce overfitting.

What is the primary purpose of padding in a convolutional layer?

To control the spatial size of the output feature map and preserve border information.

To increase the number of learnable parameters in the layer.

To reduce the computational cost of the convolution operation.

To introduce nonlinearity into the feature map.

A data science team is training a CNN on a modest dataset of 20,000 images. The training accuracy reaches 99% while validation accuracy plateaus at 85%. They suspect overfitting. Which combination of techniques is most likely to reduce the generalization gap?

Reduce the training set size and train for more epochs

Increase the number of convolutional filters and remove dropout

Add dropout layers and apply random horizontal flips and rotations during training

Replace ReLU with linear activations in hidden layers

A company needs to automatically detect and localize multiple objects in real-time video streams from security cameras. They have a large annotated dataset with bounding boxes. They want to use a pretrained CNN as a backbone. Which architecture is most appropriate for this task?

A single-stage object detector like YOLO or SSD with a pretrained backbone

An autoencoder for dimensionality reduction

A fully convolutional network without any detection head

A standard image classification CNN like VGG16 with a softmax output

How does increasing the stride of a convolutional layer affect the output feature map?

It has no effect on the spatial dimensions.

It increases the spatial dimensions of the output.

It decreases the spatial dimensions of the output.

It increases the number of channels in the output.

GIAC Machine Learning Engineer

Domain 3Objective 2

Convolutional Neural Networks GMLE Practice Questions (Page 8)

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 40 practice questions to prepare you well beyond it. (estimate)

40questions here

8free pages

10concepts

Questions 36–40

36application · medium
A deep CNN is training slowly and the activations in the hidden layers have widely varying scales across batches. The team suspects internal covariate shift. Which technique should they apply?
Select an answer first
37foundation · easy
What is the primary purpose of padding in a convolutional layer?
Select an answer first
38application · medium
A data science team is training a CNN on a modest dataset of 20,000 images. The training accuracy reaches 99% while validation accuracy plateaus at 85%. They suspect overfitting. Which combination of techniques is most likely to reduce the generalization gap?
Select an answer first
39application · medium
A company needs to automatically detect and localize multiple objects in real-time video streams from security cameras. They have a large annotated dataset with bounding boxes. They want to use a pretrained CNN as a backbone. Which architecture is most appropriate for this task?
Select an answer first
40foundation · easy
How does increasing the stride of a convolutional layer affect the output feature map?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

No more pagesBack 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.