What does a convolutional layer do in a CNN?

• Extracts features from the input image

• Reduces the dimensionality of the input

• Combines multiple feature maps

• Adds regularization to the model

What is the main purpose of a pooling layer?

• To increase the size of feature maps

• To reduce the computational cost and overfitting

• To perform element-wise multiplication

• To normalize the feature maps

Which of the following is true about max pooling?

• It averages pixel values across a region

• It selects the maximum value in a region of the feature map

• It applies a convolutional operation to input data

• It reduces the number of filters in a layer

Why are smaller filters (e.g., 3x3) preferred in modern CNN architectures?

• They reduce computation and capture local patterns

• They enhance overfitting

• They reduce vanishing gradients

• They increase spatial dimensions

What is the primary role of a CNN?

• To learn hierarchical spatial features from input images

• To predict sequential data

• To minimize the dimensionality of input data

• To solve overfitting issues

Which of the following is NOT a part of a standard CNN architecture?

• Convolutional Layer

• Pooling Layer

• Activation Layer

• Recurrence Layer

What is the input to a CNN model typically?

• Flattened vector

• Raw image data

• Audio signal

• One-hot encoded labels

What does padding do in a convolutional layer?

• Reduces the spatial dimensions of the input

• Adds zeros around the edges of the input

• Prevents overfitting

• Combines multiple feature maps

What does stride control in a CNN?

• The size of the filters

• How far the filter moves across the input

• The pooling layer's activation

• The depth of feature maps

In CNNs, increasing stride results in:

• Smaller output dimensions

• Larger output dimensions

• More overfitting

• Higher accuracy