Examen Semestre 1 et 2 : Big Data Analytics et Deep Learning

8. Laquelle des r ponses suivantes est consid r e comme le principal avantage de spark par rapport

Mapreduce ?

A. Spark supporte scala

B. Spark peut ex cuter des calculs en m moire

C. Spark supporte les RDD

D. MapReduce supporte le calcul distribu

9. Lequel des langages suivants nest pas support par Apache spark

A. Java

B. Scala

C. Python

D. Ruby

10. Lequel des composants suivants ne fait pas partie de l cosyst me de Spark ?

A. Spark core

B. Spark Name-Node

C. Spark GraphX

D. Spark MLlib

11. Laquelle des r ponses suivantes est lune des responsabilit s de Spark Driver ?

A. Il est un nSud travailleur (worker node) responsable des processus individuels dun job de

spark.

B. Il stocke les m tadonn es sur le job en cours mais il ne contient pas les donn es r elles.

C. Il est responsable de lallocation des ressources aux job lanc s.

D. Il stocke les r sultats des calculs en m moire, cache ou sur le disque.

12. Laquelle des r ponses suivantes nest pas une caract ristique valide dun RDD ?

A. L valuation paresseuse

B. La tol rance aux pannes

C. Limmuabilit

D. L volutivit

13. Etant donn la commande suivante ex cut e avec pyspark, quel est le type de donn es r sultant ?

A. Un objet FileInputStream

B. Un dataframe ou chaque l ment est une ligne depuis le fichier

C. Un pair RDD ou la cl est le chemin de fichier et la valeur du contenu du fichier

D. Un RDD ou chaque l ment est une ligne depuis le fichier.

14. Laquelle des r ponses ci-dessous nest pas une op ration valide support e par un RDD ?

A. Regroup()

B. Count()

C. parallelize()

D. filter()

3 / 8

15. Etant donn le fragment de code suivant, quel est le r sultat du programme ?

A. Le code ne sera pas compil

B. [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]

C. Le code affichera les num ros dans un ordre non d termin

D. Le code sera compil mais il naffichera rien

Partie II : Deep Learning

16. Which one of these plots represents a RELU activation function?

A. Figure n 1

B. Figure n 2

a.

C. Figure n 3

D. Figure n 4

17. What does a neuron compute?

A. A neuron computes an activation function followed by a linear function (z= Wx+b)

B. A neuron computes a function g that scales the input x linearly (Wx+b)

4 / 8

C. A neuron computes the mean of all features before applying the output to an activation function

D. A neuron computes a linear function (z= Wx+b) followed by an activation function

18. You are building a neural network where it gets input from the previous layer as well as from itself.

Which of the following architecture has feedback connections?

A. Recurrent Neural network

B. Convolutional Neural Network

C. Restricted Boltzmann Machine

D. None of these

19. Batch Normalization is helpful because?

A. It normalizes (changes) all the input before sending it to the next layer

B. It returns back the normalized mean and standard deviation of weights

C. It is a very efficient backpropagation technique

D. None of these

20. Suppose img is a (32,32,3) array, representing a 32x32 image with 3 color channels red, green and blue.

How do you reshape this into a column vector?

A. X = img.reshape((3,32*32))

B. X = img.reshape((1,3232,3))

C. X = img.reshape((32323,1))

D. X = img.reshape((32*32,3))

21. In a neural network, which of the following techniques is used to deal with overfitting?

A. Dropout

B. Regularization

C. Batch Normalization

D. All of these

22. You are building a binary classifier for recognizing cucumbers (y=1) vs. watermelons (y=0). Which one

of these activation functions would you recommend using for the output layer?

A. Relu

B. Leaky Relu

C. Sigmoid

D. Tanh

5 / 8

23. Among the following, which one is not an " hyperparameter"?

A. Learning rate

B. Number of iterations

C. Weight matrices W

D. Number of layers L in the neural network

24. Consider the following neural network, how many layers does this network have?

A. The number of layers L is 3. The number of hidden layers is 3.

B. The number of layers L is 4. The number of hidden layers is 3.

C. The number of layers L is 4. The number of hidden layers is 4

D. The number of layers L is 5. The number of hidden layers is 4.

25. In a neural network, knowing the weight and bias of each neuron is the most important step. If you can

somehow get the correct value of weight and bias for each neuron, you can approximate any function.

What would be the best way to approach this?

A. Assign random values and pray to God they are correct.

B. Search every possible combination of weights and biases till you get the best value.

C. Iteratively check that after assigning a value how far you are from the best values, and slightly

change the assigned values to make them better

D. None of these

26. What are the steps for using a gradient descent algorithm?

1. Calculate error between the actual value and the predicted value

2. Reiterate until you find the best weights of network

3. Pass an input through the network and get values from output layer

4. Initialize random weight and bias

5. Go to each neuron which contributes to the error and change its respective

values to reduce the error

A. 1, 2, 3, 4, 5

B. 5, 4, 3, 2, 1

C. 3, 2, 1, 5, 4

D. 4, 3, 1, 5, 2

6 / 8

27. In a neural network, knowing the weight and bias of each neuron is the most important step. If you can

somehow get the correct value of weight and bias for each neuron, you can approximate any function.

What would be the best way to approach this?

A. Assign random values and pray to God they are correct.

B. Search every possible combination of weights and biases till you get the best value.

C. Iteratively check that after assigning a value how far you are from the best values, and slightly

change the assigned values to make them better

D. None of these

28. Which of the following gives non-linearity to a neural network?

A. Stochastic Gradient Descent

B. Rectified Linear Unit

C. Convolution function

D. None of the above

29. What is autoencoder?

A. A neural network that codes itself

B. A neural network that decodes itself

C. An autoencoder is a type of artificial neural network used to learn efficient data codings in an

unsupervised manner.

D. An autoencoder is a type of artificial neural network used to learn efficient data codings in a

supervised manner.

30. Which of the following is an application of autoencoders?

A. Feature learning

B. Dimensionality reduction

C. Information retrieval

D. All the above

31. Consider the following statements regarding Artificial Neural Networks (ANN) and convolutional

Neural Networks (CNN):

1. There are sparse connections between inputs and outputs between two

consecutive layers in CNN.

2. CNNs can be used only for image data

3.

Parameters are shared between output neurons in CNN layer

4. Both CNNs and ANNs can take image data as input

Which of the above statements are TRUE?

A. 1 and 2

B. 1 and 4

C. 1, 3 and 4

D. 2,3 and 4

7 / 8

32. What will be the size of the output after the following operations?

Input size = [227 x 227 x 3]

Filter size = [11 x 11 x 3]

Stride = 4

2 x2 Max-pooling with stride of 2

A. [54 x 54]

B. [55 x 55]

C. [216 x 2016]

D. [68 x 68]

33. Pooling layers are used to accomplish which of the following?

A. To progressively reduce the spatial size of the representation.

B. To reduce the number of epochs while ensuring best accuracy.

C. To ovoid local minima.

D. To always select maximum value over pooling region.

Answer 34-36 for the CNN given below:

The whole network is composed of Conv layers that perform 3 x 3 convolutions with stride 2 and

padding is valid (P=1). POOL layers perform 2 x 2 max pooling with stride 2 (and no padding). Number

of filters in the Conv layers and number of neurons in fully connected layers are shown in brackets.

34. The output size after pool1, pool2 are:

A. [32 x 32 x 128], [5 x 5 x 64]

B. [31 x 31 x 128], [3 x 3 x 64]

C. [32 x 32 x 128], [3 x 3 x 64]

D. [31 x 31 x 128], [5 x 5 x 64]

35. Number of parameters till pool2 including bias are:

A. 89186

B. 297042

C. 75648

D. 147584

36. Total Number of parameters from pool2 layer till the output layer including bias are

A. 855818

B. 262400

C. 590848

D. 2570

37. Which of the following is true for most CNN architectures?

A. Size of input (height and width) decreases, while depth increases

B. Fully connected layers in the first few layers

C. Multiple convolutional layers followed by pooling layers

D. A and C

8 / 8