Convolutional Neural Networks GMLE Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   During CNN training, what is the correct order of the following steps in one iteration?

   • ABackpropagation, forward propagation, loss computation, weight update
   • BLoss computation, forward propagation, backpropagation, weight update
   • CForward propagation, loss computation, backpropagation, weight update
   • DWeight update, forward propagation, loss computation, backpropagation
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   A company needs to segment tumors in 3D MRI scans. They have a limited number of annotated volumes and require pixel-level masks. They are considering using a pretrained 2D CNN for feature extraction. Which approach is most appropriate?

   • AUse a 2D object detection model like YOLO on each slice
   • BUse a 3D U-Net architecture trained from scratch with heavy data augmentation
   • CApply a pretrained 2D CNN slice-by-slice and combine the results
   • DUse a pretrained image classification model to classify each voxel independently
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   What is the role of backpropagation in training a CNN?

   • AIt generates new training data by applying transformations to existing images.
   • BIt initializes the weights of the network to small random values.
   • CIt reduces the spatial dimensions of feature maps to lower computational cost.
   • DIt computes the gradient of the loss with respect to each weight in the network.
   Reveal Solution Discussion Select an answer first
5. 9expert · hard

   A team has a pretrained VGG16 model trained on ImageNet. They want to adapt it to classify grayscale medical images with a very small dataset (2,000 images). They also have limited computational resources. They plan to replace the input layer to accept single-channel images. Which approach is most appropriate?

   • ADiscard the pretrained weights and train the VGG16 architecture from scratch on the grayscale images
   • BModify the first convolutional layer to accept 1 channel and randomly initialize its weights, then fine-tune the entire network
   • CConvert grayscale images to 3-channel by replicating the single channel, freeze the convolutional base, and train a new classifier head
   • DUse the pretrained model as a fixed feature extractor by passing grayscale images through the RGB input layer without modification
   Reveal Solution Discussion Select an answer first
6. 10foundation · easy

   In a typical convolutional neural network used for image classification, which sequence correctly describes the main role of the core layer types?

   • AConvolutional layers extract hierarchical features, pooling layers reduce spatial dimensions, and fully connected layers perform classification.
   • BPooling layers extract features, convolutional layers reduce spatial dimensions, and fully connected layers perform classification.
   • CConvolutional layers perform classification, fully connected layers extract features, and pooling layers reduce spatial dimensions.
   • DFully connected layers extract features, convolutional layers reduce spatial dimensions, and pooling layers perform classification.
   Reveal Solution Discussion Select an answer first