CNN Convolutional Layer Operations 5

Question 1

What does the term 'feature map' refer to in the context of a convolutional layer output?

Accepted Answer

A 2D grid of activations produced by applying one filter to the input

Answer

A feature map (or activation map) is the 2D spatial output produced by convolving one filter across the entire input.

Question 2

During backpropagation through a convolutional layer, what operation is performed to compute gradients with respect to the input?

Accepted Answer

Full convolution of the upstream gradient with the flipped filter

Answer

The gradient with respect to the input is computed via convolution of the upstream gradient with the filter rotated 180°, which is equivalent to a full convolution.

Question 3

What is the computational complexity (in multiply-accumulate operations) for one convolutional layer given output H×W, K filters, and filter C×Fh×Fw?

Accepted Answer

H × W × K × C × Fh × Fw

Answer

For each of the H×W output positions and K filters, C×Fh×Fw multiply-accumulate operations are performed, giving H×W×K×C×Fh×Fw total.

Question 4

Why is local connectivity in convolutional layers advantageous over full connectivity for image data?

Accepted Answer

Nearby pixels are more correlated, so local filters capture meaningful spatial patterns with fewer parameters

Answer

Images have strong local spatial correlations, and local filters exploit this structure efficiently while greatly reducing the parameter count.

Question 5

In a CNN, what is the difference between 'valid' and 'full' convolution padding modes?

Accepted Answer

Valid produces smaller outputs with no padding; full pads enough so every input pixel is the center at least once

Answer

Valid mode uses no padding (output shrinks), while full mode pads enough for the filter to slide over every input element as a center, making output larger than input.

CNN - Convolutional Neural Networks Practice Test

CNN - Convolutional Neural Networks Practice Test

CNN Convolutional Layer Operations 5