【中英】【吴恩达课后测验】Course 4 -卷积神经网络 - 第二周测验

参考1：卷积神经网络Quiz2
参考2：quiz.md

上一篇：【课程4 - 第一周编程作业】※※※※※ 【回到目录】※※※※※下一篇：【第二周编程作业】

1. Question 1

在典型的卷积神经网络中，随着网络的深度增加，你能看到的现象是？

【】 $n_H$ 和 $n_W$ 增加，同时 $n_C$ 减少。
【】 $n_H$ 和 $n_W$ 减少，同时 $n_C$ 也减少。
【】 $n_H$ 和 $n_W$ 增加，同时 $n_C$ 也增加。
【★】 $n_H$ 和 $n_W$ 减少，同时 $n_C$ 增加。

2. Question 2

在典型的卷积神经网络中，你能看到的是？

【★】多个卷积层后面跟着的是一个池化层。
【】多个池化层后面跟着的是一个卷积层。
【★】全连接层（FC）位于最后的几层。
【】全连接层（FC）位于开始的几层。

3. Question 3

为了构建一个非常深的网络，我们经常在卷积层使用“valid”的填充，只使用池化层来缩小激活值的宽/高度，否则的话就会使得输入迅速的变小。

【】 True
【★】 False

博主注：我们经常使用“SAME”的padding方式。

4. Question 4

我们使用普通的网络结构来训练一个很深的网络，要使得网络适应一个很复杂的功能（比如增加层数)，总会有更低的训练误差。

【】 True
【★】 False

博主注：在没有残差的普通神经网络中，理论上是误差越来越低的，但是实际上是随着网络层数的加深，先减小再增加；在有残差的ResNet中，即使网络再深，训练误差都会随着网络层数的加深逐渐减小。

5. Question 5

下面计算残差(ResNet)块的公式中，横线上应该分别填什么？

a^{[l + 2]} = g (W^{[l + 2]} g (W^{[l + 1]} a^{l} + b^{[l + 1]}) + b^{[l + 2]} + \underline{?}) + \underline{?}

$a^{[l+2]} = g(W^{[l+2]}g(W^{[l+1]}a^l + b^{[l+1]}) + b^{[l+2]} + \underline{\quad ? \quad } \text{ }) + \underline{\quad ? \quad}$

【】分别是 $0$ 与 $z^{[l+1]}$ 。
【★】分别是 $a^{[l]}$ 与 $0$ 。
【】分别是 $z^{[l]}$ 与 $a^{[l]}$ 。
【】分别是 $0$ 与 $a^{[l]}$ 。

博主注：推导一下公式就好了。

\begin{aligned} (1.0) & a^{[l + 2]} & = g (z^{[l + 2]} + a^{[l]}) \\ (1.1) & = g (W^{[l + 2]} \times a^{[l + 1]} + b^{[l + 2]} + a^{[l]}) \\ (1.2) & = g (W^{[l + 2]} \times g (z^{[l + 1]}) + b^{[l + 2]} + a^{[l]}) \\ (1.3) & = g (W^{[l + 2]} \times g (W^{[l + 1]} \times a^{[l]} + b^{[l + 1]}) + b^{[l + 2]} + \underline{a^{[l]}}) + \underline{0} \end{aligned}

$\begin{align} a^{[l+2]} &= g(z^{[l+2]} + a^{[l]})\tag{1.0} \\ &= g(W^{[l+2]} \times a^{[l+1]} + b^{[l+2]} + a^{[l]}) \tag{1.1}\\ &= g(W^{[l+2]} \times g(z^{[l+1]}) + b^{[l+2]} + a^{[l]}) \tag{1.2}\\ &= g(W^{[l+2]} \times g(W^{[l+1]}\times a^{[l]} + b^{[l+1]}) + b^{[l+2]} + \underline{\quad a^{[l]} \quad}) + \underline{\quad 0 \quad} \tag{1.3} \\ \end{align}$

6. Question 6

关于残差网络下面哪个（些）说法是正确的？

【】使用跳越连接能够对反向传播的梯度下降有益且能够帮你对更深的网络进行训练。
【★】跳跃连接计算输入的复杂的非线性函数以传递到网络中的更深层。
【】有L层的残差网络一共有 $L^2$ 种跳跃连接的顺序。
【★】跳跃连接能够使得网络轻松地学习残差块类的输入输出间的身份映射。

博主注：请参考这里

7. Question 7

假设你的输入的维度为64x64x16，单个1x1的卷积过滤器含有多少个参数（包括偏差）？

【】 2
【★】 17
【】 4097
【】 1

8. Question 8

假设你有一个维度为 $n_H \times n_W \times n_C$ 的卷积输入，下面哪个说法是正确的（假设卷积层为1x1，步伐为1，padding为0）？

【★】你能够使用1x1的卷积层来减少 $n_C$ ，但是不能减少 $n_H$ 、 $n_W$
【★】你可以使用池化层减少 $n_H$ 、 $n_W$ ，但是不能减少 $n_C$
【】你可以使用一个1x1的卷积层来减少 $n_H$ 、 $n_W$ 和 $n_C$ .
【】你可以使用池化层减少 $n_H$ 、 $n_W$ 和 $n_C$ .

9. Question 9

关于 Inception 网络下面哪些说法是正确的

【】 Inception 网络包含了各种网络的体系结构（类似于随机删除节点模式，它会在每一步中随机选择网络的结构），因此它具有随机删除节点的正则化效应。
【★】 Inception 块通常使用1x1的卷积来减少输入卷积的大小，然后再使用3x3和5x5的卷积。
【★】一个inception 块允许网络使用1x1, 3x3, 5x5 的和卷积个池化层的组合。
【】通过叠加inception块的方式让inception 网络更深不会损害训练集的表现。

10. Question 10

下面哪些是使用卷积网络的开源实现（包含模型/权值）的常见原因？

【】为一个计算机视觉任务训练的模型通常可以用来数据扩充，即使对于不同的计算机视觉任务也是如此。
【★】为一个计算机视觉任务训练的参数通常对其他计算机视觉任务的预训练是有用的。
【★】使用获得计算机视觉竞赛奖项的相同的技术，广泛应用于实际部署。
【★】使用开源实现可以很简单的来实现复杂的卷积结构。

1. Question 1

Which of the following do you typically see as you move to deeper layers in a ConvNet?

[ ] $n_H$ and $n_W$ increases, while $n_C$ decreases
[ ] $n_H$ and $n_W$ decreases, while $n_C$ also decreases
[ ] $n_H$ and $n_W$ increases, while $n_C$ also increases
[x] $n_H$ and $n_W$ decrease, while $n_C$ increases

2. Question 2

Which of the following do you typically see in a ConvNet? (Check all that apply.)

[x] Multiple CONV layers followed by a POOL layer
[ ] Multiple POOL layers followed by a CONV layer
[x] FC layers in the last few layers
[ ] FC layers in the first few layers

3. Question 3

In order to be able to build very deep networks, we usually only use pooling layers to downsize the height/width of the activation volumes while convolutions are used with “valid” padding. Otherwise, we would downsize the input of the model too quickly.

[ ] True
[x] False

小傻瓜

4. Question 4

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
[x] False

5. Question 5

The following equation captures the computation in a ResNet block. What goes into the two blanks(?) above?

$a^{[l+2]}=g(W^{[l+2]}g(W^{[l+1]}a^{l}+b^{[l+1]})+b^{[l+2]}+?)+?$

[ ] 0 and $z^{[l+1]}$ , respectively
[x] $a^{[l]}$ and 0, respectively
[ ] $z^{[l]}$ and $a^{[l]}$ , respectively
[ ] 0 and $a^{[l]}$ , respectively

$a^{[l+2]}=g(z^{[l+2]}+a^{[l]})=g(W^{[l+2]}Z^{[l+1]}+b^{[l+2]}+a^{[l]})=g(W^{[l+2]}g(W^{[l+1]}a^{[l]}+b^{[l+1]})+b^{[l+2]}+a^{[l]})$

6. Question 6

Which ones of the following statements on Residual Networks are true? (Check all that apply.)

[ ] Using a skip-connection helps the gradient to backpropagate and thus helps you to train deeper networks
[x] The skip-connections compute a complex non-linear function of the input to pass to a deeper layer in the network.
[ ] A ResNet with L layers would have on the order of $L^{2}$ skip connections in total.
[x] The skip-connection makes it easy for the network to learn an identity mapping between the input and the output within the ResNet block.

7. Question 7

Suppose you have an input volume of dimension 64x64x16. How many parameters would a single 1x1 convolutional filter have (including the bias)?

[ ] 2
[ ] 17
[x] 4097
[ ] 1

64x64x1 + 1 = 4097

8. Question 8

Suppose you have an input volume of dimension $n_{H}\times n_{W}\times n_{C}$ . Which of the following statements you agree with? (Assume that “1x1 convolutional layer” below always uses a stride of 1 and no padding.)

[ ] You can use a 1x1 convolutional layer to reduce $n_C$ but not $n_H$ , $n_W$ .
[x] You can use a pooling layer to reduce $n_H$ , $n_W$ , but not $n_C$ .
[x] You can use a 1x1 convolutional layer to reduce $n_H$ , $n_W$ , and $n_C$ .
[ ] You can use a pooling layer to reduce $n_H$ , $n_W$ , and $n_C$ .

9. Question 9

Which ones of the following statements on Inception Networks are true? (Check all that apply.)

[ ] Inception networks incorporates a variety of network architectures (similar to dropout, which randomly chooses a network architecture on each step) and thus has a similar regularizing effect as dropout.
[x] Inception blocks usually use 1x1 convolutions to reduce the input data volume’s size before applying 3x3 and 5x5 convolutions.
[x] A single inception block allows the network to use a combination of 1x1, 3x3, 5x5 convolutions and pooling.
[x] Making an inception network deeper (by stacking more inception blocks together) should not hurt training set performance.

10. Question 10

Which of the following are common reasons for using open-source implementations of ConvNets (both the model and/or weights)? Check all that apply.

[ ] A model trained for one computer vision task can usually be used to perform data augmentation even for a different computer vision task.
[x] Parameters trained for one computer vision task are often useful as pretraining for other computer vision tasks.
[x] 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.
[x] It is a convenient way to get working an implementation of a complex ConvNet architecture.

Please refer to http://blog.csdn.net/koala_tree/article/details/78531398 and http://blog.csdn.net/nichengwuxiao/article/details/78709725 for further information.