ML Chapter 07

What is the main advantage of using CNNs for image data?

• They reduce memory usage
• They preserve spatial structure
• They convert images to text
• They remove noise automatically

Which of the following is a typical CNN layer sequence?

• Pool → ReLU → Conv
• ReLU → Conv → Softmax
• Conv → ReLU → Pooling
• Affine → Pool → Softmax

What is the main operation performed by a convolution layer?

• Average values
• Apply filters to detect patterns
• Randomize input
• Resize the image

What does padding do in a convolutional layer?

• Reduces the number of filters
• Converts images to grayscale
• Adds zeros around the input to preserve dimensions
• Resets weights to zero

What is the role of stride in convolution?

• It sets the filter shape
• It determines how far the filter moves
• It scales the image
• It normalizes the output

What is the output of a convolutional operation called?

• Bias field
• Loss map
• Feature map
• Pool layer

What does a pooling layer do?

• Adds noise
• Multiplies the feature maps
• Reduces spatial dimensions
• Changes the number of channels

Which type of pooling is most commonly used?

• Sum pooling
• Mean pooling
• Max pooling
• Sigmoid pooling

What stays the same after pooling?

• Image size
• Number of channels
• Number of weights
• Total data size

Why is `im2col` used in CNN implementation?

• To convert a matrix into an image
• To make convolution faster using matrix multiplication
• To visualize data
• To normalize features