吳恩達深度學習筆記 course4 week2 測驗

答案僅供參考,非標准答案,歡迎交流

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

$n_HnH and n_WnW increases, while n_CnC decreases$

$n_HnH and n_WnW increases, while n_CnC also increases$

$n_HnH and n_WnW decrease, while n_CnC increases √$

$n_HnH and n_WnW decreases, while n_CnC also decreases$

第 2 個問題

1
point

2. 第 2 個問題

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

Multiple CONV layers followed by a POOL layer

Multiple POOL layers followed by a CONV layer

FC layers in the last few layers

FC layers in the first few layers

第 3 個問題

1
point

3. 第 3 個問題

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

False √

第 4 個問題

1
point

4. 第 4 個問題

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 √

第 5 個問題

1
point

5. 第 5 個問題

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

0 and $a^{[l]}a[l], respectively$

$z^{[l+1]}z[l+1], respectively$

$z^{[l]}z[l] and a^{[l]}a[l], respectively$

$a^{[l]}a[l] and 0, respectively √$

第 6 個問題

1
point

6. 第 6 個問題

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

The skip-connection makes it easy for the network to learn an identity mapping between the input and the output within the ResNet block.

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 $L^2L2 skip connections in total.$

第 7 個問題

1
point

7. 第 7 個問題

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

17 √

4097

2

1

第 8 個問題

1
point

8. 第 8 個問題

Suppose you have an input volume of dimension $n_HnH x n_WnW x n_CnC. Which of the following statements you agree with? (Assume that “1x1 convolutional layer” below always uses a stride of 1 and no padding.)$

You can use a 1x1 convolutional layer to reduce $n_HnH, n_WnW, and n_CnC.$

You can use a 1x1 convolutional layer to reduce $n_CnC but not n_HnH, n_WnW.$

You can use a pooling layer to reduce $n_HnH, n_WnW, and n_CnC.$

You can use a pooling layer to reduce $n_HnH, n_WnW, but not n_CnC.$

第 9 個問題

1
point

9. 第 9 個問題

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

Inception blocks usually use 1x1 convolutions to reduce the input data volume’s size before applying 3x3 and 5x5 convolutions.

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.

A single inception block allows the network to use a combination of 1x1, 3x3, 5x5 convolutions and pooling.

Making an inception network deeper (by stacking more inception blocks together) should not hurt training set performance.

第 10 個問題

1
point

10. 第 10 個問題

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

Parameters trained for one computer vision task are often useful as pretraining for other computer vision tasks.

It is a convenient way to get working an implementation of a complex ConvNet architecture.

A model trained for one computer vision task can usually be used to perform data augmentation even for a different computer vision task.

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.

中文版源自:https://blog.csdn.net/u013733326/article/details/80166080

--------------------------------------------------------------------中文版---------------------------------------------------------------------

1. Question 1

在典型的卷積神經網絡中，隨着網絡的深度增加，你能看到的現象是？

【】 $n_{H}$
【】 $n_{H}$
【】 $n_{H}$
【★】 $n_{H}$

2. Question 2

在典型的卷積神經網絡中，你能看到的是？

【★】多個卷積層后面跟着的是一個池化層。
【】多個池化層后面跟着的是一個卷積層。
【★】全連接層（FC）位於最后的幾層。
【】全連接層（FC）位於開始的幾層。

3. Question 3

為了構建一個非常深的網絡，我們經常在卷積層使用“valid”的填充，只使用池化層來縮小激活值的寬/高度，否則的話就會使得輸入迅速的變小。

【】 True
【★】 False

博主注：我們經常使用“SAME”的padding方式。

4. Question 4

我們使用普通的網絡結構來訓練一個很深的網絡，要使得網絡適應一個很復雜的功能（比如增加層數)，總會有更低的訓練誤差。

【】 True
【★】 False

博主注：在沒有殘差的普通神經網絡中，理論上是誤差越來越低的，但是實際上是隨着網絡層數的加深，先減小再增加；在有殘差的ResNet中，即使網絡再深，訓練誤差都會隨着網絡層數的加深逐漸減小。

5. Question 5

下面計算殘差(ResNet)塊的公式中，橫線上應該分別填什么？

a [l + 2] = g (W [l + 2] g (W [l + 1] a l + b [l + 1]) +

【】分別是 $0$
【★】分別是 $a^{[l]}$
【】分別是 $z^{[l]}$
【】分別是 $0$

博主注：推導一下公式就好了。

a [l + 2] = g (z [l + 2] + a [l]) = g

6. Question 6

關於殘差網絡下面哪個（些）說法是正確的？

【】使用跳越連接能夠對反向傳播的梯度下降有益且能夠幫你對更深的網絡進行訓練。
【★】跳躍連接計算輸入的復雜的非線性函數以傳遞到網絡中的更深層。
【】有L層的殘差網絡一共有 $L^{2}$
【★】跳躍連接能夠使得網絡輕松地學習殘差塊類的輸入輸出間的身份映射。

博主注：請參考這里

7. Question 7

假設你的輸入的維度為64x64x16，單個1x1的卷積過濾器含有多少個參數（包括偏差）？

【】 2
【★】 17
【】 4097
【】 1

8. Question 8

假設你有一個維度為 $n_{H} \times n_{W} \times n_{C}$

【★】你能夠使用1x1的卷積層來減少 $n_{C}$
【★】你可以使用池化層減少 $n_{H}$
【】你可以使用一個1x1的卷積層來減少 $n_{H}$
【】你可以使用池化層減少 $n_{H}$

9. Question 9

關於 Inception 網絡下面哪些說法是正確的

【】 Inception 網絡包含了各種網絡的體系結構（類似於隨機刪除節點模式，它會在每一步中隨機選擇網絡的結構），因此它具有隨機刪除節點的正則化效應。
【★】 Inception 塊通常使用1x1的卷積來減少輸入卷積的大小，然后再使用3x3和5x5的卷積。
【★】一個inception 塊允許網絡使用1x1, 3x3, 5x5 的和卷積個池化層的組合。
【】通過疊加inception塊的方式讓inception 網絡更深不會損害訓練集的表現。

10. Question 10

下面哪些是使用卷積網絡的開源實現（包含模型/權值）的常見原因？

【】為一個計算機視覺任務訓練的模型通常可以用來數據擴充，即使對於不同的計算機視覺任務也是如此。
【★】為一個計算機視覺任務訓練的參數通常對其他計算機視覺任務的預訓練是有用的。
【】使用獲得計算機視覺競賽獎項的相同的技術，廣泛應用於實際部署。
【★】使用開源實現可以很簡單的來實現復雜的卷積結構。