CNN Convolutional Layer Operations 4 — Questions and Answers
Question 1: What is dilated (atrous) convolution, and what advantage does it provide?
- A convolution with gaps between filter elements, expanding receptive field without adding parameters (Correct answer)
- A convolution that uses fractional strides
- A convolution applied only to dilated images
- A convolution with larger filters and more parameters
Correct answer: A convolution with gaps between filter elements, expanding receptive field without adding parameters
Dilation inserts zeros between filter elements (dilation rate > 1), covering a larger area of the input with the same number of parameters.
Question 2: What is the effect of applying batch normalization after a convolutional layer before the activation function?
- Normalizes activations to have zero mean and unit variance, stabilizing training (Correct answer)
- Removes the need for filters entirely
- Replaces the bias term with a learned mean
- Doubles the number of parameters in the layer
Correct answer: Normalizes activations to have zero mean and unit variance, stabilizing training
Batch normalization standardizes pre-activation values across the mini-batch, reducing internal covariate shift and enabling higher learning rates.
Question 3: In a transposed convolution (sometimes called deconvolution), what is the typical use case in CNNs?
- Spatial upsampling to increase feature map resolution (Correct answer)
- Reducing depth of feature maps
- Applying convolution in the frequency domain
- Reversing gradient flow during backpropagation
Correct answer: Spatial upsampling to increase feature map resolution
Transposed convolutions learn to upsample feature maps and are used in decoder networks, GANs, and semantic segmentation architectures.
Question 4: Given input 28×28×1, filter size 5×5, padding=2, stride=1, how many parameters are in one filter (including bias)?
- 26 (Correct answer)
- 25
- 27
- 50
Correct answer: 26
One filter has 5×5×1 = 25 weights plus 1 bias = 26 parameters.
Question 5: Which term describes sharing the same filter weights across all spatial positions in a convolutional layer?
- Weight sharing (parameter sharing) (Correct answer)
- Sparse connectivity
- Equivariance
- Translation invariance
Correct answer: Weight sharing (parameter sharing)
Weight sharing means the same filter is applied at every position, dramatically reducing parameters compared to a fully connected layer.
Question 6: What property of convolutional layers makes them well-suited for detecting the same feature regardless of its position in the image?
- Translation equivariance (Correct answer)
- Rotational invariance
- Scale invariance
- Depth invariance
Correct answer: Translation equivariance
Translation equivariance means that if the input shifts, the output feature map shifts by the same amount, allowing position-independent detection.
Question 7: How does grouped convolution (used in ResNeXt) reduce computational cost compared to standard convolution?
- It splits input channels into groups and applies independent filters per group, reducing multiplications (Correct answer)
- It removes all bias terms
- It applies filters only to the border pixels
- It uses stride=2 automatically
Correct answer: It splits input channels into groups and applies independent filters per group, reducing multiplications
By restricting each filter to convolve with only a subset of input channels (one group), grouped convolution reduces FLOPs by the group factor.
CNN Convolutional Layer Operations 4
What is dilated (atrous) convolution, and what advantage does it provide?