Neural Network Convolutional Neural Networks 2

Question 1

What is depthwise separable convolution and what is its main advantage?

Accepted Answer

A factored convolution applied depthwise then pointwise, dramatically reducing parameters and computation

Answer

A convolution that processes depth information only

Answer

A convolution that separates positive and negative activations

Answer

A 3D convolution for volumetric data

Question 2

What is feature map visualization and why is it used in CNN analysis?

Accepted Answer

Displaying the intermediate activations to understand what features each layer detects

Answer

Visualizing the training loss over time

Answer

Plotting the distribution of filter weights

Answer

Showing the confusion matrix of predictions

Question 3

What is transfer learning in the context of CNNs?

Accepted Answer

Reusing pretrained CNN weights as a starting point for a new task

Answer

Transferring a model from one GPU to another

Answer

Converting a CNN model to run on mobile devices

Answer

Sharing weights between encoder and decoder networks

Question 4

In CNN-based image classification, what does the global average pooling layer do before the final classifier?

Accepted Answer

Averages each entire feature map to a single value, reducing spatial dimensions to 1×1

Answer

Applies max pooling across all feature maps simultaneously

Answer

Pools gradients globally for stability

Answer

Concatenates all feature maps into a single vector

Question 5

What does data augmentation accomplish during CNN training?

Accepted Answer

It artificially expands the training set by applying random transformations to existing samples

Answer

It increases the model's depth automatically

Answer

It normalizes pixel values to zero mean

Answer

It generates synthetic labels for unlabeled data

Question 6

What is Grad-CAM and how is it used?

Accepted Answer

A visualization technique using gradients to highlight image regions important for a prediction

Answer

A gradient-based optimizer for CNNs

Answer

A type of convolutional layer with learned pooling

Answer

A camera-based data collection tool for training