Back to course
Next page
Interactive Quiz
Test your knowledge!
1
What main problem do residual connections help solve in deep neural networks?
A
They prevent overfitting by reducing the number of parameters.
B
They facilitate optimization by allowing the network to more easily learn functions close to the identity function.
C
They increase the size of input images to improve resolution.
D
They completely replace convolutions with pooling operations.
2
In a classic ResNet block, how is the residual connection adapted when the number of output filters differs from the number of input filters?
A
Only a pooling operation is applied.
B
A 1x1 convolution is used to adjust the number of channels.
C
Input filters are duplicated to match the number of output filters.
D
An additional activation function is used before the sum.
3
What is the main difference between a basic ResNet block and a bottleneck block?
A
The bottleneck block uses fewer 1x1 convolutions than the basic block.
B
The bottleneck block contains more convolutions but is faster due to 1x1 convolutions.
C
The bottleneck block does not contain residual connections.
D
The bottleneck block is only used for shallow networks.
4
Why is the activation function (such as ReLU) applied after adding the residual part in a ResNet block?
A
To consider the entire block as a standalone layer with its own non-linearity.
B
To normalize values before addition.
C
To reduce the dimension of tensors.
D
To prevent the sum from always being zero.
5
What effect do residual connections have on the loss function during training, according to studies?
A
They make the loss function more complex and non-convex.
B
They smooth the loss function, thereby facilitating optimization.
C
They increase the optimal learning rate.
D
They do not modify the shape of the loss function.
Score: 0/5
Score: 0/5