Back to course
Next page
Interactive Quiz
Test your knowledge!
1
What is the major drawback of using a fully connected neural network directly on a large image such as 224x224x3?
A
It results in a significant loss of information.
B
The number of parameters becomes too high, making training expensive.
C
The network cannot process color images.
D
Fully connected networks are incompatible with images.
2
What is the main function of convolutional layers in a convolutional neural network?
A
Reduce the image dimension by removing pixels.
B
Share weights to detect local patterns while limiting the number of parameters.
C
Convert a color image to a grayscale image.
D
Apply a nonlinear transformation to the pixels.
3
In a 2D convolutional layer, what does the kernel_size parameter specify?
A
The number of filters used in the layer.
B
The size of the filter's stride.
C
The size of the convolution filter, for example, 3 means a 3x3 filter.
D
The number of output channels.
4
What is the main effect of a stride greater than 1 in a convolutional layer?
A
Increase the spatial resolution of the output.
B
Reduce the spatial resolution of the output by subsampling.
C
Increase the number of output channels.
D
Add padding pixels around the image.
5
What is the role of padding in a convolutional layer?
A
Add values at the borders to allow the filter to be applied uniformly across all pixels.
B
Increase the depth of the filters.
C
Reduce the size of the input image.
D
Apply a nonlinear activation after convolution.
6
What is the main difference between MaxPooling and AveragePooling in a convolutional network?
A
MaxPooling takes the maximum value of a group of pixels, AveragePooling takes the average.
B
MaxPooling increases the size of the feature map, AveragePooling decreases it.
C
MaxPooling is parameterized, AveragePooling has no parameters.
D
MaxPooling applies a convolution, AveragePooling applies an activation.
7
Why does stacking multiple convolutional layers increase the receptive field of a network?
A
Because each layer increases the size of the filters.
B
Because successive layers combine local interactions to cover a larger area.
C
Because padding is increased at each layer.
D
Because the stride is always set to 1.
8
In a U-Net architecture for segmentation, what is the main purpose of the connections between the encoder and the decoder?
A
Reduce the size of the output images.
B
Preserve precise local information while allowing global abstraction.
C
Increase the depth of the network.
D
Enable the use of 1D convolution.
9
Which layer is generally used to increase the spatial resolution in a convolutional network, particularly in the decoder part of a U-Net?
A
A MaxPooling layer.
B
A Fully Connected layer.
C
A transposed convolution layer (ConvTranspose).
D
A Batch Normalization layer.
10
What advantage have convolutional networks shown on datasets like MNIST compared to traditional fully connected networks?
A
Greater ease in processing grayscale images but with more parameters.
B
Similar accuracy but with many more parameters.
C
Better accuracy with far fewer parameters.
D
Better accuracy but only on color images.
Score: 0/10
Score: 0/10