Deep Convolutional Models
- Which of the following do you typically see in a ConvNet? (Check all that apply.)
- Multiple CONV layers followed by a POOL layer
- FC layers in the last few layers
- Multiple POOL layers followed by a CONV layer
- FC layers in the first few layers
- In LeNet - 5 we can see that as we get into deeper networks the number of channels increases while the height and width of the volume decreases. True/False?
- True
- False
- Training a deeper network (for example, adding additional layers to the network) allows the network to fit more complex functions and thus almost always results in lower training error. For this question, assume we’re referring to “plain” networks.
- True
- False
解析:Resnets are here to help us train very deep neural networks.
- **The following equation captures the computation in a ResNet block. What goes into the two blanks above? **
- a [ l ] a^{[l]} a[l] and 0, respectively
- 0 and a [ l ] a^{[l]} a[l], respectively
- 0 and z [ l + 1 ] ] z^{[l+1]]} z[l+1]], respectively
- z [ l ] z^{[l]} z[l] and a [ l ] a^{[l]} a[l], respectively
- Which ones of the following statements on Residual Networks are true? (Check all that apply.)
- A ResNet with L layers would have on the order of L 2 L2 L2 skip connections in total.
- The skip-connections compute a complex non-linear function of the input to pass to a deeper layer in the network.
This should not be selected
This is false, skip connections make it easy for the model to learn an identity mapping, not a complex non-linear function.
- The skip-connection makes it easy for the network to learn an identity mapping between the input and the output within the ResNet block.
- Using a skip-connection helps the gradient to backpropagate and thus helps you to train deeper networks
- Suppose you have an input volume of dimension n H × n W × n C n_H \times n_W \times n_C nH×nW×nC.Which of the following statements do you agree with? (Assume that the “1x1 convolutional layer” below always uses a stride of 1 and no padding.)
- You can use a 1x1 convolutional layer to reduce n C n_C nC but not n H n_H nH and n W n_W nW.
Yes, a 1x1 convolutional layer with a small number of filters is going to reduce nCnC but will keep the dimensions nHnH and nWnW
- You can use a 2D pooling layer to reduce n H n_H nH and n W n_W nW, but not n C n_C nC.
- You can use a 2D pooling layer to reduce n H n_H nH and n W n_W nW, and n C n_C nC.
- You can use a 1x1 convolutional layer to reduce n H n_H nH, n W n_W nW, and n C n_C nC.
- Which of the following are true about the inception Network? (Check all that apply)
- Inception blocks allow the use of a combination of 1x1, 3x3, 5x5 convolutions and pooling by stacking up all the activations resulting from each type of layer.
The use of several different types of layers and stacking up the results to get a single volume is at the heart of the inception network.
- Making an inception network deeper won’t hurt the training set performance.
- One problem with simply stacking up several layers is the computational cost of it.
That is why the bottleneck layer is used to reduce the computational cost.
- Inception blocks allow the use of a combination of 1x1, 3x3, 5x5 convolutions, and pooling by applying one layer after the other.
- Which of the following are common reasons for using open-source implementations of ConvNets (both the model and/or weights)? Check all that apply.
- It is a convenient way to get working with an implementation of a complex ConvNet architecture.
- Parameters trained for one computer vision task are often useful as pre-training for other computer vision tasks.
- The same techniques for winning computer vision competitions, such as using multiple crops at test time, are widely used in practical deployments (or production system deployments) of ConvNets.
Using multiple crops at test time is not common.
- A model trained for one computer vision task can usually be used to perform data augmentation for a different computer vision task
- Which of the following are true about Depth wise-separable convolutions? (Choose all that apply)
- The result has always the same number of channels n c n_c nc as the input.
- They are just a combination of a normal convolution and a bottleneck layer.
- They combine depthwise convolutions with pointwise convolutions.
this combination is what we call depth wise separable convolutions.
- They have a lower computational cost than normal convolutions.
Yes, as seen in the lectures the use of the depthwise and pointwise convolution reduces the computational cost significantly.
- Fill in the missing dimensions shown in the image below (marked W, Y, Z).
- W = 5, Y = 30, Z = 20
- W = 30, Y = 20, Z =20
- W = 5, Y = 20, Z = 5
- W = 30, Y = 30, Z = 5