A team is training a CNN on a dataset of satellite images to classify land use. The dataset has class imbalance, with one class representing 90% of the samples. The model achieves high accuracy but poor recall on the minority class. Which combination of techniques is most appropriate to address this issue?

Increase the number of filters in the convolutional layers to improve model capacity.

Apply dropout after every convolutional layer to regularize the model.

Use data augmentation on the minority class and adjust the class weights in the loss function.

Use a larger batch size to stabilize training.

What is the main benefit of using transfer learning with a pretrained CNN like ResNet for a new image classification task?

It eliminates the need for any training data.

It allows the model to leverage learned features from a large dataset, reducing training time and data requirements.

It guarantees higher accuracy than training from scratch on any dataset.

It automatically adjusts the input size to match the new dataset.

During CNN training, a model's loss decreases rapidly for the first few epochs but then starts oscillating and validation loss increases. The learning rate is 0.1. Which adjustment is most likely to stabilize training?

Remove batch normalization layers

Reduce the learning rate to 0.001

Increase the batch size to the full dataset

Increase the learning rate to 0.5

A team is building a CNN to classify images where the object of interest may appear at different scales and positions. They want to achieve translation invariance while preserving enough spatial information for fine-grained classification. Which pooling strategy is most appropriate?

Use average pooling with a large kernel (e.g., 5x5) in every layer

Avoid pooling entirely and rely only on strided convolutions

Use max pooling with a moderate kernel size (e.g., 2x2) after several convolutional layers

Use global average pooling immediately after the first convolutional layer

A company needs to detect multiple objects in an image and draw bounding boxes around them. They have a pretrained CNN for image classification. What is the most efficient way to adapt it for object detection?

Add a global average pooling layer and a softmax layer for multi-label classification.

Use the pretrained CNN as a fixed feature extractor and train a support vector machine on the features.

Replace the final fully connected layer with a regression head for bounding boxes and a classification head for object classes.

Train the pretrained CNN from scratch on the object detection dataset.

GIAC Machine Learning Engineer

Domain 3Objective 2

Convolutional Neural Networks GMLE Practice Questions (Page 1)

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 1–5

1application · medium
A team is training a CNN on a dataset of satellite images to classify land use. The dataset has class imbalance, with one class representing 90% of the samples. The model achieves high accuracy but poor recall on the minority class. Which combination of techniques is most appropriate to address this issue?
Select an answer first
2foundation · easy
What is the main benefit of using transfer learning with a pretrained CNN like ResNet for a new image classification task?
Select an answer first
3application · medium
During CNN training, a model's loss decreases rapidly for the first few epochs but then starts oscillating and validation loss increases. The learning rate is 0.1. Which adjustment is most likely to stabilize training?
Select an answer first
4expert · hard
A team is building a CNN to classify images where the object of interest may appear at different scales and positions. They want to achieve translation invariance while preserving enough spatial information for fine-grained classification. Which pooling strategy is most appropriate?
Select an answer first
5application · medium
A company needs to detect multiple objects in an image and draw bounding boxes around them. They have a pretrained CNN for image classification. What is the most efficient way to adapt it for object detection?
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.