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

The learned weight matrix of a filter

Answer

The gradient map used during backpropagation

Answer

The flattened output before the fully connected layer

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

Element-wise multiplication of the filter and upstream gradient

Answer

Pooling of upstream gradients

Answer

Transposing the weight matrix and multiplying

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

H × W × K

Answer

K × Fh × Fw

Answer

H × W × C

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

Full connectivity is computationally cheaper

Answer

Local connectivity allows larger batch sizes

Answer

It removes the need for activation functions

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 pads with ones; full pads with zeros

Answer

Valid uses stride=1; full uses stride=2

Answer

There is no difference in output size

Question 6

What is the role of the activation function applied after a convolutional layer?

Accepted Answer

Introduces non-linearity so the network can learn complex feature hierarchies

Answer

Normalizes the weights of the filter

Answer

Reduces the spatial dimensions of the feature map

Answer

Converts the feature map to a scalar

Question 7

What distinguishes a convolutional layer from a locally connected layer (unshared convolution)?

Accepted Answer

Convolutional layers share filter weights across all positions; locally connected layers use different weights at each position

Answer

Locally connected layers share weights; convolutional layers do not

Answer

They are identical in parameter count

Answer

Locally connected layers cannot handle multi-channel inputs