CNN Architecture Fundamentals 2

Question 1

What is the primary purpose of padding in a convolutional layer?

Accepted Answer

To preserve spatial dimensions of the input

Answer

To reduce the number of filters

Answer

To increase the depth of the feature maps

Answer

To normalize pixel values before convolution

Question 2

In a CNN, what does 'stride' control?

Accepted Answer

How many pixels the filter moves between applications

Answer

The depth of the convolutional filter

Answer

The number of channels in the output

Answer

The learning rate during backpropagation

Question 3

Which layer type in a CNN is responsible for introducing non-linearity?

Accepted Answer

Activation layer

Answer

Pooling layer

Answer

Fully connected layer

Answer

Normalization layer

Question 4

What distinguishes a depthwise separable convolution from a standard convolution?

Accepted Answer

It applies one filter per input channel, then combines with 1×1 convolutions

Answer

It uses larger kernel sizes to capture more context

Answer

It replaces pooling layers entirely

Answer

It operates only on the fully connected layers

Question 5

In AlexNet, what was the purpose of Local Response Normalization (LRN)?

Accepted Answer

To normalize activations across neighboring feature maps for lateral inhibition

Answer

To replace dropout during training

Answer

To scale gradients during backpropagation

Answer

To reduce the spatial size of feature maps

Question 6

How does a 1×1 convolution (pointwise convolution) affect a feature map?

Accepted Answer

It changes the number of channels without altering spatial dimensions

Answer

It increases spatial resolution

Answer

It applies spatial filtering across a local region

Answer

It acts as a pooling operation

Question 7

What is the receptive field of a neuron in a CNN?

Accepted Answer

The region of the input image that influences that neuron's activation

Answer

The set of weights in its corresponding filter

Answer

The number of channels it processes

Answer

The output size of the layer it belongs to