Deep Learning Quiz- 1

In Convolutional Neural Networks (CNNs), how are kernels applied?

• a) A single kernel is reused everywhere
  

• b) Many kernels are applied in parallel
  

• c) Only diagonal kernels are used
  

• d) No kernels are applied

Why are multiple kernels used in CNNs?

• a) To reduce memory usage
  

• b) To detect different types of features
  

• c) To increase dataset size
  

• d) To perform pooling
  

What is the name of the output produced by applying a kernel?

• a) Loss function
  

• b) Gradient
  

• c) Feature map
  

• d) Epoch

What happens to multiple feature maps in a CNN layer?

• a) They are discarded
  

• b) They are stacked together
  

• c) Only the first one is kept
  

• d) They replace the input image directly

Why is stacking feature maps important?

• a) It compresses the dataset
  

• b) It allows learning multiple features at once
  

• c) It removes noisy data
  

• d) It avoids overfitting

What is the key difference between standard convolution and CNN convolution?

• a) Standard uses pooling, CNN does not
  

• b) Standard uses one kernel, CNN uses many
  

• c) Standard is non-linear, CNN is linear
  

• d) Standard applies dropout, CNN does not

What is the main difference between a convolutional layer and a locally connected layer?

• a) Locally connected layers use pooling, but convolution does not
  

• b) Convolution layers use weight sharing, locally connected layers do not
  

• c) Convolution is nonlinear, locally connected is linear
  

• d) Convolution has fewer parameters than locally connected layers

Why are locally connected layers useful in face detection?

• a) Faces have no structure
  

• b) Features appear in fixed regions (eyes at top, mouth at bottom)
  

• c) They use fewer parameters than CNNs
  

• d) They completely replace convolution layers

Which of the following is a disadvantage of locally connected layers compared to convolutional layers?

• a) They require fewer parameters
  

• b) They are more memory-efficient
  

• c) They require more parameters (no weight sharing)
  

• d) They do not work with images

Why are multiple channels useful in CNNs?

• a) They allow combining different types of information at each location
  

• b) They reduce the size

• of the dataset
  c) They remove noise automatically
  

• d) They make images black and white