Hcia Ai Dump File

1 Dump Detail
1. Page 2 to 91 ============== Dumps

2. Page 2 to 91 ============== Dumps
3. Page 2 to 91 ============== Exam#1 Dumps
2
1.Which of the following is not included in the recurrent neural network usage scenario?
A. Machine
translation
B.Speechrecogn on
C. Image style migration
D. Text generation
Answer: ABCD
2. Which of the following about the gradient descent is incorrect?
A. Random gradient descent is a commonly used one in gradient descent.
B. Gradient descent includes random gradient descent and batch gradient descent.
C. The gradient descent algorithm 1s fast and reliable [Right Answers}
D. Random gradient descent is one of the commonly used optimization algorithms in deep learning
algorithms.
Answer: C
3. Which of the following can improve the computational efficiency of the neural network model? (Multjple
choice)
A. TPU
B. GPU
C. FPGA
D. Large-scale distributed cluster
Answer: A B C D
4. From the perspective of technical architecture Al The classification of chips includes?
A. FPGA
B.CPU
C.GPU
D.ASIC
3
Answer: A B C D
4
5. The for loop statement in the Python language can iterate through the items in any sequence.
A. True
B. False
Answer: A
6. TensorFlow in keras The built-in optimizer of the module has?
A. Adam
B. SGD
C. Adaboost
D. Adadelta
Answer: A B D
7. Traditional machine learning and deep learning are the core technologies of artificial intelligence.
There is a slight difference in the engineering process. The following steps. What you don't need to do
in deep learning is:
A. Model evaluation
B. Feature engineering
C. Data cleaning
D. Model building
Answer: B
8. Where should the labeled data be placed in the confrontation generation network?
A. As the output value of the generated model
B. As the input value of the discriminant model
C. As the output value of the discriminant model
D. As input value for generative model
Answer: C
5
9. What are the scenarios or industries that are suitable for using Python? (Multiple Choice)
A. Artificial intelligence
B. web development
C. Game development
D Hardware development
Answer: A BC
10. On Huawei Cloud El Which of the following can be Al Technology is integrated into the application
scenarios of all walks of life Al The advantages of technology to improve efficiency and improve
experience.
A. El Agent
B. OBS Object storage service
C. Cloud database
D. El Big data service
Answer: A
11. TensorFlow It is an end-to-end open source platform for machine learning and deep learning.
A. TRUE
B. FALSE
Answer: A
12. Atlas 200 DK Yiteng 310 The main application of the chip is?
A. Model reasoning
B. Build model
C. Training model
Answer: A
13. L1 with L2 Regularization is a method commonly used in traditional machine learning to reduce
generalization errors. The following is about the two. The right way is:
6
A. L1 Regularization can do feature selection
B. L1 with L2 Regularization can be used for feature selection
C. L2 Regularization can do feature selection
D. L1 with L2 Regularization cannot be used for feature selection
Answer: A
14. Linear algebra is a branch of algebra that deals with linear problems.
The linear relationship is that the relationship between exponential objects is expressed in ( ) times form?
A. 1
B. 2
C. 3
D. 4
Answer: A
15. Functions are well-organized, non-reusable code segments used to implement a single, or associated
Function.
A. True
B. False
Answer: B
16. Global gradient descent algorithm, stochastic gradient descent algorithm and batch gradient descent
algorithm are all gradient descent algorithms. The following is wrong about its advantages and
disadvantages.
A. The global gradient algorithm can find the minimum value of the loss function
B. Batch gradient algorithm can solve the local minimum problem
C. Stochastic gradient algorithm can find the minimum value of the loss function
D. The convergence process of the global gradient algorithm is time-consuming
Answer: C
7
17. Deep learning neural network training requires a lot of matrix calculations. Generally , we need to use
hardware to enable the computer to have parallel computing capabilities. The following hardware devices
can provide parallel computing capabilities.:
A. Motherboard
B. RAM
C. GPU
D.CPU
Answer: C
18. Artificial intelligence at this stage is still in the weak artificial intelligence stage.
A. TRUE
B. FALSE
Answer: A
19. In polynomial regression, there is a square term in the formula of the model, so it is not linear.
A. TRUE
B. FALSE
Answer: B
20. Voice recognition refers to the recognition of audio data as text data.
A. TRUE
B. FALSE
Answer: A
21. In TensorFlow, data is represented in the form of tensors and calculation charts.
A. True
B. False
Answer: B
+
8
22. What are the advantages of Python? (Multiple choice)
A. Simple
B. Free
C. High-level language
D. Rich library
Answer: A B C D
23. Pytorch Which company launched it first?
A. Baidu
B. Google
C. Facebook
D. Huawei
Answer: C
24. The Python language can use multiple statements on the same line, separated by commas".".
A. True
B. False
Answer: B
25. What are the application scenarios for the break statement in the Python language? (Multiple Choice)
A. Any Python statement
B. while loop statement
C. for loop statement
D. Nested loop statement
Answer: BCD
26. The following description of machine learning is correct?
A. Deep learning is a branch of machine learning
B. Deep learning and machine learning are mutually inclusive

9
C. Deep learning and machine learning are both artificial intelligence but have no relationship to each other
D. None of the above
Answer: A
27. Tensorflow Operations and Computation Graph are not - run in the Session
A. True
B. False
Answer: B
28. What is wrong description of the normal distribution?
A. In natural phenomena and social phenomena, many random variables obey or approximate a normal
distribution.
B. The normal distribution lakes the maximum value at the mean
C. The larger the standard deviation the steeper the normal d1stribuhon curve
D. The larger the standard deviation, the slower the normal distribution curve.
Answer: C
29. Which of the following are the characteristics of TensorFlow? (Multiple choice)
A. Open source
B. Convenience
C. Mature
D. Flexible
Answer: A B C D
30. Traffic agent can achieve () Hour full time/Global traffic awareness?
A. 7x24
B. 5x
24
C.5x12
10
D.7x12
11
Answer: A
31. In order for a machine to be intelligent, it must be knowledgeable. Therefore, there is a research field in
artificial intelligence. which mainly studies how computers automatically acquire knowledge
and skills to achieve self-impr011emenL What is the branch of this research called?
A. Expert system
B. Machine learning
C. Neural Network
D. Natural language processing
Answer: B
32. tf.keras.datasets Can view keras The built-in data set.
A. TRUE
B. FALSE
Answer: A
33. When we describe the house, we usually use the attributes such as residential area, house type,
decoration type, etc. If we use plain With Bayes as a model, we assume that there is no relationship
between attributes.
A. TRUE
B. FALSE
Answer: A
34. The following are the description of the distribution function. distribution law. and density function of
random variables.
which is incorrect?
A. Distribution law can only describe the value rule of discrete random variables
B. The density function can only describe the value rule of continuous random variables
C. Discrete random variables have no d1slribulion function

12
D. Distribution function describes the value rule of random variables
Answer: C
35. During the two classification process, we can set any category as a positive example.
A. TRUE
B. FALSE
Answer: A
36. The constituent elements of the Python identifier include: numbers, letters, and underscores.
A. True
B. False
Answer: A
37. Which is the correct description of the python creation function? (Multiple choice}
A. The function created starts with the def keyword followed by the function name and parentheses.
B. The parameters need to be placed in parentheses
C. The function content starts with a colon and needs lo be indented
D. Return the result with return and the function ends.
Answer: A B C D
38. There are many commercial applications for machine learning services. What are the main business
scenarios covered? (Multiple Choice)
A. Financial product recommendation
B. Predictive maintenance
C. Telecom customer retention
D. Retailer grouping
Answer: A B C D
39. Which of the following are the cloud services provided by Huawei Cloud El Visual Cognition? (Multiple
13
choice)
A. Text recognition
B. Face recognition
C. Image recognition
D. Content detection
E. Image processing
Answer: A B C D E
40. GBDT Compared with the random forest algorithm, which of the following statements is wrong?
A. GBDT Algorithm is easier to underfit than random forest
B. Random forest is calculated in parallel, while GBDT Can't
C. GBDT Algorithm is easier to overfit than random forest
D. GBDT And random forest are built on CART Based on the tree
Answer: C
41. What is wrong about the image content review service?
A. politics Test results for sensitive persons involved in politics
B. terrorism Test results for political violence
C. confidence Represents confidence, range 0-100
D. labe1 Label representing each test result
Answer: C
42. Which of the following about the dictionary in Python is correct? (Multiple Choice)
A. Each key and its corresponding value need to be separated by ":
B. Separate the different key-value with ,
C. The entire dictionary is included in the {}
D. The keys of the dictionary are unique and the data type is uniform
Answer: A B C D
14
43. Which of the following about the description of the number of rows and columns for the determinant is
correct?
A. The number of rows is greater than the number of columns
B. The number of rows is equal to the number of columns
C. The number of rows is less than the number of columns
D. The number of rows has no relationship with the number of columns.
Answer: B
44. What people say about neural networks is wrong?
A. As the number of hidden layers of the neural network increases, the classification ability of the model
gradually weakens
B. The limitation of a single-layer perceptron is that it cannot solve the XOR problem
C. The feedforward neural network can be represented by a directed acyclic graph
D. There is no connection between neurons in the same layer of the feedforward neural network
Answer: A
45. Which of the following descriptions of the depth feedforward network is correct?
A. Deep feedforward network 1s a kind of neural network.
B. The deep feedforward network has only one hidden layer.
C. The unit on the hidden layer of the deep feedforward network will have countless.
D. Deep feedforward network 1s used to deal with linear problems.
Answer: A
46. The following options do not belong to the scope of Huawei's full-stack solution are?
A. Application enable
B. Edge computing
C. Open source framework
D. Chip enable
Answer:C
15
47. Cut off 2019 At the end of the year,TensorFlow The released versions are?
A. 1
B.3
C. 2
D. 0.1
Answer: ACD
48. Which of the following is not a specific technology of artificial intelligence?
A. Knowledge map
B. Machine translation
C. Riemann geometry
D. Semantic understanding
Answer: C
49. Which of the following capabilities does the genetic knowledge map have?
A. Auxiliary case diagnosis
B. Disease prediction and diagnosis
C. Gene test report generation
D. Entity query
Answer: ACD
50. In the gradient descent algorithm, which of the following algorithms is the most confusing algorithm for
the trajectory on the loss function surface?
A. SGD
B. BGD
C. MGD
D. MBGD
Answer: A
16
51. The following about the standard RNN Model, the correct statement is?
A. There is no one-to-one model structure
B. Do not consider the time direction when backpropagating
C. There is no many-to-many model structure
D. There will be a problem of attenuation of long-term transmission and memory information
Answer: D
52. The test error will keep getting smaller as the complexity of the model increases.
A. TRUE
B. FALSE
Answer: B
53. TensorFlow 2.0 The way to view tensor dimensions in is?
A. dimens
B. dtype
C. ndim
D. device
Answer: C
54. In the environment preparation for invoking HUAWEI CLOUD services, which of the following does not
require confirmation of installation?
A. requests
B. setuptools
C. websocket-client
D. Jieba
Answer: D
55. In supervised learning," Those who are near Zhu are red and those who are near Mo are black"ls used
17
to describe which of the following models?
A. K-Means
B. SVM
C.KNN
D. Neural Networks
Answer: C
56. The model composed of machine learning algorithms cannot represent the true data distribution
function on a theoretical level. Just approach it.
A. TRUE
B. FALSE
Answer: A
57. HiAI What are the advantages of mobile computing platforms?
A. Quick source code
B. Diverse tool chain
C. Complete documentation
D. plentiful API
Answer: A B C D
58. What is the incorrect relationship between neural networks and deep learning?
A. The concept of deep learning sterns from the study or artificial neural networks.
B. A neural network algorithm with multiple hidden layers is a deep learning algorithm
C. Single layer neural networks are also a type of deep learning
D. Convolutional neural networks are a type of deep learning.
Answer: C
59. What model is not a cyclic neural network?
A. RNN
18
B. LSTM
C.GBOT
D. GRU
Answer: C
60. Which of the following about the description of expectations and variances is incorrect?
A. Expectation reflects the average level of random variable values
B. The variance reflects the degree of deviation between the random variable and its mathematical
expectation
C. Expectation and variance are both numerical characteristics of random variables
D. The greater the expectation the smaller the variance
Answer: D
61. What is in the category of artificial intelligence? (Multiple choice)
A. Action
B. Perception
C. Cognition
Answer: ABC
62. Deep learning is a branch of machine learning
A. True
B. False
Answer: A
63. What is the performance of artificial intelligence in the stage of perceptual intelligence?
A. Machines begin to understand, think and make decisions like humans
B. Machines begin to calculate and transmit information just like humans
C. The machine starts to understand and understand, make judgments, and take some simple actions
19
Answer: A
64. Deep learning algorithms can be divided into supervised learning and unsupervised learning.
A. True
B. False
Answer: B
65. In random forest, what strategy is used to determine the outcome of the final ensemble model?
A. Cumulative system
B. Find the average
C. Voting system
D. Cumulative system
Answer:ABC
66. Grid search is a method of parameter adjustment.
A. TRUE
B. FALSE
Answer: A
67. Which is not a Python operator?
A. arithmetic operator
B. Reasoning operator
C. Logical operator
D. Comparison operator
Answer: B
68. What are the service categories included in Huawei Cloud El Enterprise Intelligence? (Multiple Choice)
A. El visual cognition
B. El online games
20
C. El speech semantics
D. El industry scene
Answer: AC
69. Which is wrong with the Python module time introduction method?
A. import time
B. from time import*
C. import time as t
D. from time
Answer: D
70. Which of the following data types does Tensorflow not support?
A. int8
B. ftoat32
C. double64
D. bool
Answer: C
71. The trace operation returns the sum of the diagonal elements of the matrix. Therefore, the trace of
matrix A and its transposed matrix are equal
A. True
B. False
Answer: A
72. TensorFlow 2.0 stand by GPU The acceleration is mainly manifested in the parallel operation of the
addition, subtraction, multiplication and division of the matrix accelerate.
A. TRUE
B. FALSE
Answer: A
21
73. The pooling layer in the convolution al neural network can reduce the size of the lower layer input.
Common pooling is:
A. Minimum strata
B. Product pooling layer
C. Maximum pooling layer
D. Average pooling layer
Answer: CD
74. Which descriptions are correct about python's index? (Multiple Choice)
A. Index from left to right defaults from 0
B. Index from left to right defaults from 1
C. Index from right lo left defaults from -1
D. Index from right to left defaults from 0
Answer: AC
75. Which of the following items are included in the results returned when the face search service is
successfully called?
A. Searched face similarity
B. Searched faces id
C. Searched face position
D. Searched face number
Answer: ABC
76. In a neural network, knowing the weight and deviations of each neuron is the most important step. If
you know the exact weights and deviations of neurons in some way, you can approximate any function
What is the best way to achieve this?
A. Random assignment, pray that they are correct
B. Search for a combmat1 on of weight and deviation until the best value 1s obtained
22
C. Assign an initial value to iteratively update weight by checking the difference between the best value
and the initial
D. The above is not correct
Answer: C
77. The correlation coefficient. also known as the linear correlation coefficient, is used to measure the
linear relationship between two variables, which is a real number greater than zero
A. True
B. False
Answer: B
78. In deep learning tasks, when encountering data imbalance problems, which of the following methods
can we use to solve the problem?
A. batch deletion
B. Random oversampling
C. Synthetic sampling
D. Random undersampling
Answer: BCD
79. Deep learning is different from machine learning and there are no unsupervised algorithms
A. True
B. False
Answer: B
80. Use with Atlas300 (3000) What conditions need to be checked when the accelerator card
server compiles and runs the program?
A. carry out Atlas Driver Installation
B. It has been installed CUDA Software package
C. It has been installed Cmake Compilation tool

23
D. Complete environment configuration
Answer: AC D
81. Deflated Al The logical architecture of the processor does not include which of the following options?
A.DVPP
B. GPU
C. Al Calculation engine
D. Chip system control CPU
Answer: B
82. What are the mainstream deep learning open source tools? (Multiple choice)
A. TensorFlow
B. Gaffe
C. Torch
D. Theano
E. Scikit..Jeam
Answer: A B C D
83. HUAWEI HiAI Engine Can easily combine multiple Al Ability and App integrated.
A. TRUE
B. FALSE
Answer: A
84. What are the algorithms supported by Tensorflow? (Multiple Choice}
A. GNN
B. ZNN
C. RNN
D. HUT
Answer: AD
24
85. The loss function of logistic regression is the cross-entropy loss function.
A. TRUE
B. FALSE
Answer: A
86. The following about the general form identification service returned type The field statement is correct?
A. type Representative text recognition area type
B. type for text Time represents the text recognition area
C. type Representative form type
D. type for table Time represents the form recognition area
Answer: B D
87. In a neural network based on connectionism, each node can express a specific meaning.
A. TRUE
B. FALSE
Answer: A
88. Feature is the dimension that describes the characteristics of the sample. Regarding its interpretability
in traditional machine learning and deep learning, the following statement is correct:
A. Features are interpretable in traditional machine learning, but weak in deep learning
B. Features are weak in traditional machine learning, but strong in deep learning
C. Features are weak in interpretability in traditional machine learning and deep learning
D. Features are interpretable in traditional machine learning and deep learning
Answer: A
89. Huawei's chip support HUAWEI HiAI Which module of?
A. HiAI Engine
B. HiAI Foundation
25
C. HiAI Framework
D. HiAI Service
Answer: ABD
90. When dealing with actual problems, when should machine learning be used in the following situations?
A. The data distribution itself changes over time, requiring continuous re-adaptation of the program, such
as predicting the trend of merchandise sales
B. The complexity of the rules is low, and the problem is small
C. Task rules will change over time, such as defect detection on the production line
D. The rules are very complicated or cannot be described, such as face recognition and voice recognition
Answer: AC D
91. Which of the following statement about the matrix is incorrect?
A. The arbitrary matrix is multiplied by the unit matrix and will not change.
B. The transpose of the symmetric matrix A 1s also A itself.
C. The transpose of the orthogonal matrix A is equal to the inverse of A .
D. There is no inverse matrix for the diagonal matrix.
Answer: D
92. SVM What are the commonly used kernel functions in?
A. Gaussian kernel function
B. Polynomial kernel function
C. Sigmiod Kernel function
D. Linear kernel
Answer: A B C D
93. In the process of training the neural network, our goal is to keep the loss function reduced. Which of the
following methods do we usually use to minimize the loss function?
A. Gradient descent
26
B. Dropout
C. Cross-validation
D. Regularization
Answer: A
94. The three stages of artificial intelligence include computational intelligence, () , Cognitive intelligence.
A. Weak artificial intelligence
B. Perceptual intelligence
C. Behavioral intelligence
D. Strong artificial intelligence
Answer: B
95. What are the Python language data types? (Multiple Choice)
A. numbers
B. string
C. list
D. tuple
E. dictionary
Answer: A B C D E
96. Regarding the convolutional neural network pooling layer, the following description is correct?
A. Pooling operation is realized by scanning window
B. The pooling layer can play a role in dimensionality reduction
C. Commonly used pooling methods include maximum pooling and average pooling
D. The pooled feature image becomes smaller
Answer: A B C D
97. Which of the following are included in the application scenarios ofTensorflow? (Multiple choice}
A.SpeechrecognWon
27
B. Face recognition
C. Image style changes
D. Autopilot
Answer: A B C D
98. TensorFlow2.0 of keras.preprocessing The role is?
A. keras Data processing tools
B. keras Built-in optimizer
C. keras Model deployment tool
D. Keras Generative model tool
Answer: A
99. TensorFlow2.0 The methods that can be used for tensor merging are?
A. join
B. concat
C. split
D. unstack
Answer: B
100. Which of the following is not a way for the TensorFlow program to read data?
A. Preload data
B. Feeding data
C. Read from the file
D. Write a file format reader
Answer: D
101. HUAWEI CLOUD ModelArts Is for Al Which of the following functions are in the developed one-stop
development platform Mode1Arts Can have?
A. Data governance
28
B. Al market
C. Visual workflow
D. Automatic learning
Answer: B C D
102. TensorFlow only supports Python and R language
A. True
B. False
Answer: B
103. Where is the main foothold of symbolism?
A. The foothold is in neuron networks and deep learning.
B. The foothold is reasoning, symbolic reasoning and machine reasoning.
C. The foothold is perception and action.
D. The foothold is behavior control, adaptive and evolutionary computing.
Answer: B
104. Computer vision is the study of how to make computers "see".
A. TRUE
B. FALSE
Answer: A
105. What are the numeric types of Python? (Multiple Choice)
A. int (integer type)
B. long (long integer type)
C. float (floating point type)
D. complex (complex number type)
Answer: A C D
29
106. Which of the following options is not the Tensorflow build process?
A. Building a calculation chart
B. Input tensor
C. Generate a session
D. Update weights I Right Answers)
Answer: D
107. In which year was the first introduction of "artificial intelligence"?
A. 1946
B. 1960
C.1916
D. 1956
Answer: D
108. Recurrent neural networks can capture dynamic information in serialized data.
A. TRUE
B. FALSE
Answer: A
109. Which of the following statements about supervised learning is correct?
A. Decision tree is a supervised learning
B. Supervised learning cannot use cross-validation for training
C. Supervised learning is a rule-based algorithm
D. Supervised learning can be trained without labels
Answer: A
110. Which of the following description about Python regular expression is not correct?
A. Python has added there regular expression module since version 1.5.
B. re regular expression module gives the Python language all the regular expression functionality.
30
C. re regular expressions are powerful tools for working with strings.
D. re regular expressions can only process string data not numeric data
Answer: D
111. There are a lot of data generated during the training of the neural network. What mechanism does
TensorFlow use to avoid excessive input data?
A. Client
B. feed
C. placeholder
D. fetch
Answer: C
112. In the deep learning network, the backpropagation algorithm is used to find the optimal parameters .
What rules are used in the backpropagation algorithm to obtain the derivation layer by layer?
A. Chain rule
B. Cumulative rule
C. Law of reciprocity
D. Rule of Normalization
Answer: A
113. With Atlas300 (3000)Install the accelerator card in the server DDK (Device Development Kit) Only
need to install when dependent on the environment Python3.
A. TRUE
B. FALSE
Answer: A
114. Which of the following description of the number of rows and columns of the matrix is correct?
A. The number of rows 1s greater than the number of columns
B. The number of rows 1s equal to the number of columns

31
C. The number of rows is less than the number of columns
D. The number of rows has no relationship with the number of columns.
Answer: D
115. TensorFlow It is one of the most popular deep learning frameworks.
A. TRUE
B. FALSE
Answer: A
116. With a lot of sales data but no labels, companies want to identify VIP Customer, the follo wing model
Suitable?
A. Logistic regression
B. SVM
C. K-Means
D. Hierarchical clustering
Answer: CD
117. As the following, what are the Python language design philosophy? (Multiple Choice)
A. Beautiful
B. Expensive
C. Explicit
D. Simple
Answer: ACD
118. What are the core concepts in TensorFlow? (Multiple Choice)
A. Gridding
B. Calculation Chart
C. Dot Product
D. Tensor
32
Answer: B D
119. What of the following does belong to convolution al neural network (CNN)? (Multiple Choice)
A. VGGNet
B. ResNet
C. AlexNet
D. GoogleNet
Answer: B
120. Recurrent neural network is different from convolutional neural network, it is bette r at solving the
following problems?
A. Sequence related issues
B. Image classification
C. Image detection
D. Recommended question
Answer: A
121. The TensorFlow framework does not support Windows systems.
A. True
B. False
Answer: B
122. . Which of the following is not MindSpore Features of the core architecture?
A. Automatic differentiation
B. Automatic tuning
C. Automatic coding
D. Automatic parallel
Answer: D
33
123. Which of the following activation functions are prone to vanishing gradient problems?
A.ReLU
B. Softplus
C. Tanh
D. Sigmoid
Answer: AD
124. Ce11 Provides basic modules for defining and performing calculations, Ce11 The object can be
executed directly, the following statement is wrong?
A. _init_, Initialization parameters(Parameter), Submodule(Ce11),operator(Primitive)Equal group Software
for initial verification
B. Construct, Define the execution process. In graph mode, it will be compiled into graphs for execution,
and there is no syntax restriction
C. There's some left optim Commonly used optimizers,wrap Pre-defined commonly used network
packaging functions Ce11
D. bprop (Optional), The reverse of the custom module
Answer: AC D
125. None of the following options TensorFlow2.0. The supported operators are:
A. pow
C. @
C. ^
D. II
Answer: C
126. Which of the following descriptions about Python program are correct? (Multiple choice)
A. The program consists of modules
B. Module contains statements
C. statement contains an expression

34
D. Expressions create and process objects
Answer: ABCD
127. During application development, which of the following operations is not a typical network definition?
A. Network definition
B. Network execution
C. Knowledge distillation
D. Weight initialization
Answer: C
128. Which of the following statements about passport recognition services are correct?
A. country_code Country code representing the country where the passport was issued
B. nationality Nationality of representative holder
C. passport_number Representative passport number
D. confidence Confidence information of related fields. The greater the confidence, the higher the reliability
of the corresponding field identified this time. In a statistical sense, the greater the confidence , the higher
the accuracy
Answer: A B C D
129. Which of the following options is not central to linear algebra?
A. Probability theory
B. Linear transformation
C. Matrix theory
D. Vector space
Answer: A
130. What does not belong to supervised learning?
B. Support vector machine

35
C. Decision tree
D. Principal component analysis
Answer: D
131. Al Chip is also called Al Accelerators are specially designed to handle a large number of computing
tasks in artificial intelligence applications functional module.
A. TRUE
B. FALSE
Answer: A
132. Convolutional neural networks are more suitable for image recognition problems than cyclic neural
networks.
A. True
B. False
Answer: A
133. What is the English abbreviation for Al?
A. Automatic Intelligence
B. Artifical Intelligence
C. Automatic Information
D. Artifical Information
Answer: B
134. What are the steps that are not part of the operation of a Python file object?
A. open
B. delete
C. read
D. write
E. close
36
Answer: B
135. Which of the following environments does not support to install the Tensor Flow?
A. Linux
B. Mac OS
C. Docker
D. OpenStack
Answer: D
136. Reducing the gap between the training error and the test error will result in over-fitting How to prevent
over-fitting? (Multiple Choice)
A. Cross validation
B. Integration method
C. Increase regularization
D. Feature Engineering
Answer: A B C D
137. Which of the following is not a face recognition technology?
A. Face search
B. Face comparison
C. Face Detection
D. Remake recognition
Answer: D
138. The training error will continue to decrease as the model complexity increases.
A. TRUE
B. FALSE
Answer: B
37
139. In the deep learning neural network, the perceptron is the simplest neural network. The correct
statement about its structure is:
A. There are only two hidden layers
B. Only one hidden layer
C. Its network uses Sigmoid Activation function
D. Its network uses Relu Activation function
Answer: B
140. HUAWEI CLOUD El Enable more use of corporate boundaries Al And big data services to accelerate
business development and benefit society. HUAWEI CLOUD El The service can serve the enterprise in the
following aspects?
A. Industry data
B. Industry wisdom
C. algorithm
D. Computing power
Answer: A B C D
141. PyTorch Which of the following functions does not have?
A. lnline keras
B. Support dynamic graph
C. Automatic derivative
D. GPU accelerate
Answer: D
142. Which of the following are solutions for the Huawei Cloud El industry scenario? (Multiple Choice)
A. Intelligent Logistics
B. Intelligent Water
C. Intelligent transportation
D. Intelligent Finance
38
E. Intelligent manufacturing
Answer: A C D E
143. PyTorch All existing versions support Python2 with Python3?
A. TRUE
B. FALSE
Answer: B
144. An e-commerce company has always had problems with official website stalls, poor utilization of
network resources, and difficulty in operation and maintenance. Then which of the following can the
company use El service?
A. Park Agent
B. Traffic Agent
C. Network agent
D. Industrial Agent
Answer: C
145. Which of the following are the characteristics of the Python language? {Multiple choice)
A. Explanatory
B. Process oriented
C. Object-oriented
D. Dynamic data type
Answer: ABCD
146. Not supported in the following options TensorFLow2.0 The attributes for dimensional transformation
are.
A. squeeze
B. reshape
C. gather
39
D. transpose
Answer: C
147. Which of the following are the elements of artificial intelligence? (Multiple choice)
A. Algorithm
B. Computing power
C. Data
D.Scene
Answer: A B C D
148. According to the development process of the robot, it is usually divided into three generations ,
respectively are: (Multiple Choice)
A. Teaching Reproduction Robot
B. Robot with sensation
C. Robots that will think
D. Intelligent robot
Answer: ABD
149. K Folding cross-validation refers to dividing the test data set into K Sub-data sets.
A. TRUE
B. FALSE
Answer: A
150. Which of the following is not an artificial intelligence school?
A. Symbolism
B. StatJst1caltsm
C. Behaviorism
D. Connectionism
Answer: B
40
151. Facing industry research and full scenarios Al The huge gap between applications, MindSpore
Bridging the application gap to help inclusiveness Al Of technological innovation does not include which of
the following?
A. New ways of collaboration
B. New programming language
C. New programming paradigm
D. New execution mode
Answer: B
152. Artificial intelligence is the research and development of theories, methods and application systems
used to simulate, extend and expand human intelligence Of a new technological science.
A. TRUE
B. FALSE
Answer: A
153. Which command can be checked Atlas 300 (3000) Whether the accelerator card is in place?
A. 1spci I grep'npu'
B. 1spci I grep'd100'
C. 1spci grep'atlas'
D. atlas info
Answer: B
154. What are the common types of dirty data?
A. Malformed value
B. Duplicate value
C. Logically wrong value
D. Missing value
Answer: A B C D
41
155. In the face search service, if we want to delete a certain face set, we can use this code:
firs_cliet.get_v2().get_face_set_service().delete_face_set("**"), among them"**" is to fill in the actual face
set name.
A. TRUE
B. FALSE
Answer: A
156. The Python list can be identified by'[]", and the default index of the first element from left to right is 1.
A. True
B. False
Answer: B
157. The following code was used when compiling the model:
model.compile(optimizer='Adam,loss='categorical.crossentropy',metrics=[tf.keras.metrics.accuracy]),
currently using evaluate When the method evaluates the model, which of the following indicators will be
output?
A. accuracy
B. categorical_ Loss
C. loss
D. categorical accuracy
Answer: AC
158. The Python language can use the "#" at the beginning of a single line of code for code comments .
A. True
B. False
Answer: A
159. What are the conditions form row n column matrix A and p row q column matrix B to be multiplied?
42
43
A. m*p. n*q
B. n*p
C. m=n
D. p=q
Answer: A
160. The following applications that are not part of the adversarial generation network are?
A. Text generation
B. Image generation
C. Image Identification
D. Data enhancement
Answer: C
161. In neural networks, weights are often shareD. Which of the following neural networks will share
weights?
A. sensor
B. Convolutional Neural Network
C. Fully connected neural network
D. Recurrent neural network
Answer: B D
162. Regarding backpropagation, the following statement is wrong?
A. Backpropagation can only be used in feedforward neural networks
B. Backpropagation can be combined with gradient descent algorithm to update network weights
C. Backpropagation passes through the activation function
D. Back propagation refers to the back propagation of errors through the network
Answer: A
163. The following belongs to TensorFlow2.0 The characteristic is?

44
A. Introduce Keras interface
B. Support static image mechanism
C. Support more platforms and more languages
D. Continue to be compatible Tensorflow 2.0x Related modules
Answer: AC D
164. The following is the correct difference between machine learning algorithms and traditional rule-based
methods?
A. Traditional rule-based methods, in which the rules can be explicitly clarified manually
B. Traditional rule-based methods use explicit programming to solve problems
C. The mapping relationship of the model in machine learning is automatically learned
D. The mapping relationship of the model in the machine learning institute must be implicit
Answer: ABC
165. Cyclic neural networks are more suitable for dealing with image recognition problems.
A. True
B. False
Answer: B
A. TRUE
B. FALSE
Answer: B
167. Python tuples are identified by "()" and internal elements are separated by":".
A. True
B. False
Answer: B
45
168. If a model has a large deviation on the test set and a small variance, it means that the model?
A. Overfitting
B. May be overfitting may be underfitting
C. Just fit
D. Underfitting
Answer: D
169. Jobs that are repetitive and require weak social skills are the easiest to be Al Replaced work.
A. TRUE
B. FALSE
Answer: A
170. Among the following properties TensorFlow2.0 Does not support creation tensor. The method is?
A. zeros
B. fill
C. create
D. constant
Answer: C
171. Huawei Machine learning Service MLS MLS is a one-stop platform that supports the entire process of
data analysis. Which of the following is not a feature of MLS?
A. A rich library of machine learning algorithms.
B. machine learning program is intuitive and easy to use.
C. Distributed and scalable big data computing engine.
D. Support for the R language but does not support the Python language
Answer: D
172. TensorFlow2.0 The mechanism of graphs and conversations has been cancelled in.
A. TRUE
46
B. FALSE
Answer: B
173. TensorFlow2.0 middle Keras The three main advantages of the interface are user-friendly, modular
and Combination, easy to expand.
A. TRUE
B. FALSE
Answer: A
174. In Huawei Cloud El Enterprise Intelligence, which basic platform services are included? (Multiple
Choice)
A. Machine learning
B. Deep learning
C. Graph engine
D. Batch processing
Answer: A B C D
175. About Bayesian formula- P(WIX)=P(XIW)*P(W)/P(X) What is the correct description?
A. P(WIX) is a prior probability
B. P(XIW) is a conditional probability
C. P(W) is the posterior probability
D. P(X) is the posterior probability
Answer: A
176. ModelArts Service and (). The combination of services can easily deploy the model to "end"?
A. OBS
B. OCR
C. ECS
D. HiLens
47
Answer: D
177. Which of the following neural network structures will share weights? (Multiple choice)
A. Convolutional neural network
B. Recurrent neural network
D. All of the above
Answer: AB
178. What are the commonly used activation functions? (Multiple Choice)
A. sigmoid
B. tanh
C. relu
D. danish
Answer: ABC
179. What quotation marks can the Python language use? (Multiple Choice)
A. Single quotes
B. Double quotes
C. Three quotes
D. Four quotes
Answer: ABC
180. On-Device Execution, that is, the entire image is offloaded and executed, and the computing power of
the Yiteng chip can be fully utilized, which can greatly reduce the interaction overhead, thereby increasing
the accelerator occupancy rate. On-Device The following description is wrong?
A. MindSpore Realize decentralized autonomy through adaptive graph optimization driven by gradient data
A11 Reduce, Gradient aggregation is in step, and calculation and communication are fully streamlined
B. Challenges of model execution under super chip computing power: Memory wall problems, high
48
interaction overhead, and difficulty in data supply. Partly in Host Executed, partly in Device Execution,
interaction overhead is even much greater than execution overhead, resulting in low accelerator occupancy
C. MindSpore Through the chip-oriented depth map optimization technology , the synchronization wait is
less, and the "data computing communication" is maximized. The parallelism of "trust", compared with
training performance Host Side view scheduling method is flat
D. The challenge of distributed gradient aggregation under super chip computing power:ResINet50
Single iteration 20ms Time will be generated The synchronization overhead of heart control and the
communication overhead of frequent synchronization. Traditional methods require 3 Synchronization
completed A11 Reduce, Data-driven method autonomy A11 Reduce, No control overhead
Answer: C
181. Which of the following is HUAWEI HiAI Foundation Function of the module?
A. App integrated
B. Let the service actively find users
C. Quickly convert and migrate existing models
D. According to user needs, push services at the right time and at the right time
Answer: D
182. Which description is wrong about the hyperparameter?
A. hyper parameters are parameters that set values before the algorithm begins learning.
B. Most machine learning algorithms have hyperparameters.
C. Hyperparameters cannot be modified
D. The value of the hyperparameter is not learned by the algorithm itself.
Answer: C
183. Which of the following is an image label service URI?
A. /v1.0/image/recognition
B. /v1.0/image/celebrity-recognition
C. /v1.0/image/recapture-detect
49
D. /v1.0/image/tagging
Answer: D
184. CPU Mainly through increasing instructions, increasing the number of cores, etc. to improve Al
performance.
A. TRUE
B. FALSE
Answer: A
185. Faced with the challenge of achieving efficient distributed training for ultra-large-scale models,
MindSpore is handled as?
A. Automatic parallel
B. Serial
C. Manual parallel
Answer: A
186. Which of the following conditions is not a condition that n-fold bernoulli trials needs to meet?
A. Each test was repeated under the same conditions.
B. There are only two possible outcomes for each trial, i.e. event A occurs and event A does not occur.
C. Each test itself is obeying normal distribution
D. The results of each trial are independent of each other.
Answer: C
187. Information theory is a branch of applied mathematics. The main research is to quantify how much
information a signal contains. Who is the first to propose information theory?
A. Minsky
B. Turing
C. Simon
D. Shannon
50
Answer: D
188. Generating a confrontation network is like a game system. The generator generates fake samples,
and the discriminator judges whether it is true or false. Our ideal result is?
A. The generator produces roughly the same samples
B. The discriminator efficiently distinguishes the true and false samples generated by the generator
C. The discriminator cannot distinguish the true and false samples generated by the generator
D. The generator produces different samples
Answer: C
189. Which of the following options belong to Tensorflow concept? (Multiple Choice)
A. Tensor
B. Variables
C. Placeholder
D. Operation
E. Conversation
Answer: A B C D E
190. Which of the following are the activation functions of deep learning algorithms?
A. Sigmoid
B. ReLU
C. Tanh
D. Sin
Answer: ABC
191. in TensorFlow2.0 in tf.contrib The method can continue to be used.
A. TRUE
B. FALSE
Answer: B
51
192. twenty two.Among machine learning algorithms, which of the following is not unsupervised learning?
A. GMM
B. Xgboost
C. Clustering
D. Association rules
Answer: B
193. Which is not a deep learning algorithm?
A. Setr-encoder
B. Convolutional neural networks
C. Recurrent neural networks
D. Support vector machine
Answer: C
194. After the data has completed the feature engineering operation, in the process of constructing the
model, which of the following options is not a step in the decision tree construction process?
A. Pruning
B. Feature selection
C. Data cleaning
D. Decision tree generation
Answer: C
195. TensorFlow2.0 of keras.metrics The methods that are not included in the built-in indicators are?
A. Accuracy
B. Recall
C.Sum
D. Mean
Answer: C
52
196. Which of the following is not a module in the Tensorflow library?
A. tf.nn
B. tf. layers
C. tf boost
D. tf.contrib
Answer: C
197. A scalar k is multiplied by matrix A equal to k and each of the numbers in A is multiplied.
A. True
B. False
Answer: A
198. In the classic convolutional neural network model,Softmax What hidden layer does the function follow?
A. Convolutional layer
B. Pooling layer
C. Fully connected layer
D. All of the above
Answer: C
199. Tensorflow is the second generation of artificial intelligence learning system developed by Google
based on().
A. DistBelief
B. PaleyFunction
C. ConvexOne
D. Infinity
Answer: A
200. Root The user cannot install Atlas300 (3000) Install the accelerator card on the server DDK (Device
53
Development Kit), Can only be installed by ordinary users.
A. TRUE
B. FALSE
Answer: A
201. The naive Bayes algorithm does not require independent and identical distribution among sample
features.
A. TRUE
B. FALSE
Answer: B
202. Machine Learning Service is a data mining and analytics platform service that helps users quickly
discover data patterns and build predictive models through machine learning techniques and deploy them
as predictive analytics solutions
A. True
B. False
Answer: A
203. Regularization is an important and effective technique for reducing generalization errors in traditional
machine learning. The following techniques are regular. The technology is:
A. L1 Regularization
B. L2 Regularization
C. Dropout
D. Momentum optimizer
Answer: ABC
204. Which of the following schools does the neural network study belong to?
A. Symbolism
B. Connectionism
54
C. Behaviorism
Answer: B
205. In deep learning, if the neural network has a large number of layers, the problem of gradient
disappearance is more likely to occur. Strictly speaking, in which of the following links does the sample
disappearance problem occur?
A. Backpropagation update parameters
B. Forward the dream and update
C. Backpropagation calculation result
D. Forward calculation results
Answer: A
206. Which of the following options is not the- session mode used by Tensorflow?
A. Explicitly call the session to generate function
B. Explicitly call the session to close function
C. Through the Python context manager
D. Multiple POST queries
Answer: D
207. The label predicted by the regression algorithm is?
A. Self-deformation
B. Discrete
C. Strain type
D. Continuous
Answer: D
208. Huawei Al The whole scenario includes public cloud, private cloud, various edge computing, loT
industry terminals and consumer. The deployment environment of terminal, edge, and cloud for fee-based
55
terminals.
A. TRUE
B. FALSE
Answer: A
209. The following options belong to keras under estimator The encapsulation methods are?
A. Assessment
B. training
C. prediction
D. Output model
Answer: ABC
210. As shown below, which are the following matrix A characteristic value? (Multiple choice)
A. 2
B. -2
C. 4
D.4
Answer: AD
211. Which of the following are the topics of speech processing research?
A. Speech processing
B. Voiceprint recognition
C. Speech Recognition
D. Wake up
Answer: BCD
212. Artificial intelligence is currently suitable for scenarios with I' known environment, clear objectives, and
56
predictable action±. Deep learning in the fields of image recognition, speech recognition, translation, etc ,
artificial intelligence basically has human recognition ability. even surpassing human beings Based on
these capabilities. it has been applied to many scenes such as medical care and public safety However,
there is still a lack of reasoning and cognition
A. True
B. False
Answer: A
213. What research areas does artificial intelligence have? (Multiple choice)
A. Natural language processing
B. Computer Vision
C. Machine learning
D. Speech recognition
Answer: A B C D
214. The meaning of artificial intelligence was first proposed by a scientist in 1950, and at the same time a
test model of machine intelligence was proposed Who is this scientist?
A. Minsky
B. Zade
C. Turing
D. Von Neumann
Answer: C
215. The Python dictionary is identified by "{}'', and the internal data consists of the key and
its corresponding value.
A. True
B. False
Answer: A
57
216. The time interval in the Python language is a Hoating-point fraction in seconds
A. True
B. False
Answer: A
217. Tensor Yes MindSpore The storage component of the data.
A. TRUE
B. FALSE
Answer: A
218. Which of the following descriptions about the Recurrent Neural Network (RNN) is correct?
A. Can be used to process sequence data.
B. Cannot process variable length sequence data.
C. Unlike convolutional neural networks. parameters of RNN cannot be shared
D. The units above the hidden layer are not associated With each other.
Answer: A
219. Regular term can also be added to logistic regression to avoid overfitting.
A. TRUE
B. FALSE
Answer: A
220. Tensorflow supports multi-TPU cluster computing
A. True
B. False
Answer: A
221. If the deep learning neural network has the problem of gradient disappearance or gradient explosion,
our common solution is.

58
A. Gradient shear
B. Random undersampling
C. use Relu Activation function
D. Regularization
Answer: AC D
222. What are the commonly used loss functions? (Multiple choice)
A. Mean variance
B. Sigmoid cross entropy
C. Soft max cross entropy
D. Sparse cross entropy
E. Weighted Sigmoid Cross Entropy
Answer: A B C D E
223. Which of the following does not belong to long-term memory LSTM (long Short·Term Memory)
network architecture?
A. Memory door
B. Forget the door
C. Input door
D. Output door
Answer: A
224. The number of rows and columns of the three matrices A, B. and C is 3 rows and 2 columns, 2 rows.
3 columns. 3 rows and 3 columns. which of the following operations is meaningful?
A.AC
B. BC
C.A+B
D. AB-BC
Answer: B
59
225. Which of the following statements about universal form recognition services are correct?
A. rows Represents the line information occupied by the text block, the number is from O Start, list form
B. colums Represents the column information occupied by the text block, the number is from O Start,
list form
C. The incoming image data needs to go through base64 coding
D. words Representative text block recognition result
Answer: A B C D
226. How many spaces does PEP 8 stipulate that Python's first fine indentation needs to indent?
A. 1
B.2
C. 4
D. 8
Answer: C
227. Generative confrontation networks are currently widely useD. The following scenarios can use this
network?
A. Data enhancement
B. Semantic segmentation
C. Information retrieval
D. Image generation
Answer: A B C D
228. The Python dictionary is widely identified by"{}". and the internal data consists of the key and its
corresponding value
A. True
B. False
Answer: A
60
229. Which method is not supported in TensorFlow to define variables?
A. Random number
B. Constant
C. Calculated from the initial values of other variables
D. null
Answer: D
230. Python script execution mode includes interactive mode and script mode
A. True
B. False
Answer: A
231. Loss function and model function are the same thing.
A. TRUE
B. FALSE
Answer: B
232. Which of the following description of the validation set is wrong?
A. The verification set can coincide with the test set.
B. The test set can coincide with the training set
C. The subset used to pick hyperparameters is called a validation set
D. Typically 80% of the training data 1s used for training and 20% 1s used for verification.
Answer: A
233. TensorFlow The data type used is?
A. Scalar
B. Vector
C. Tensor
61
D. Matrix
Answer: C
234. Vector group a1=(1, 1, 1), a2=(0,2.5), a3=(1.3,6), which of the following options is correct?
A. Linear correlation
B. Linear independence
C.aHa2+a3=0
D.2a1+a2+a3=0
Answer: A
235. What is not the optimization method in deep learning?
A. Random gradient descent
B. Back propagation algorithm
C. Principal component analysis
D. Momentum
Answer: C
236. Which of the following are Al Subfield?
A. Machine learning
B. Computer vision
Answer: A B C D
237. In a convolutional neural network, different layers have different functions. Which of the following
layers can play the role of dimensionality reduction?
A. Input layer
B. Fully connected layer
C. Convolutional layer
62
D. Pooling layer
Answer: BCD
238. The number of hidden layers of the deep learning neural network has a certain impact on the
performance of the network. The following statement about its impact is correct.:
A. The number of hidden layers is appropriately reduced, and the resolution ability of the neural network
remains unchanged
B. Increase the number of hidden layers appropriately, the stronger the resolution ability of the neural
network
C. The number of hidden layers is appropriately reduced, the stronger the resolution ability of the neural
network is
D. The number of hidden layers increases appropriately, the weaker the resolution ability of the neural
network
Answer: B
239. Atlas accelerate Al What processor is used for inference?
A. Different 910 processor
B. Different 310 processor
C. GPU
D. FPGA
Answer: B
240. When using TensorFlow2. 0 of keras When building a neural network with an interface, the network
needs to be compiled. Which of the following methods should be used to work?
A. compile
B. write
C. join
D. fit
Answer: A
63
241. The main computing resources included in the Da Vinci architecture computing unit are?
A. Vector calculation unit
B. Scalar Computing Unit
C. Tensor computing unit
D. Matrix calculation unit
Answer: AB D
242. In what year did Huawei officially provide services in the form of cloud services, and combined with
more partners to provide richer artificial intelligence practices?
A.2002
B.2013
C.2015
D.2017
Answer: D
243. Which of the following does the convolution al neural network structure not include?
B. Pooling layer
C. Loop layer
D. Fully connected layer
Answer: C
244. What are the regularizations in deep learning? (Multiple Choice)
A. L1 norm. L2 norm
B. Data set enhancement
C. Integration method
D. Dropout
Answer: A B C D
64
245. Al Chips are divided into business applications and can be divided into?
A. training
B. GPU
C. Model building
D. reasoning
Answer: AD
246. The number or matrix is stored in the tensor
A. True
B. False
Answer: B
247. HUAWEI CLOUD El There are many types of agents according to different industries and application
scenarios. Currently, Huawei Cloud El The agent has?
A. Industrial Agent
B. Traffic Agent
C. Park Agent
D. Car intelligence
Answer: A B C D
248. Linear regression in 3 The fitted surface in the dimension above dimension is?
A. Curved surface
B. flat
C. Hyperplane
D. Hypersurface
Answer: C
249. The commonly used functions for mathematical operations in Python are basically in lhe math module
65
and the cmath module.
A. True
B. False
Answer: A
250. Vector is a number.
A. True
B. False
Answer: B
251. " print" in Python 3 must be used with "()"
A. True
B. False
Answer: A
252. There are many types of neural networks in deep learning. The following neural network information is
one-way propagation:
A. LSTM
C. Recurrent neural network
D. GRU
Answer: B
253. Which of the following statements about overflow and underflow in numerical calculations is correct?
(Multiple choice)
A. Underflow occurs when the number close to zero Is rounded to zero
B. Overflow occurs when the number close to zero 1s rounded to zero
C. Overflow occurs when a large number of levels are approximated as 1P or - 1P
D. Underflow occurs when a large number of levels are approximated as 1P or - 1P

66
Answer: AC
254. The image data of the passport recognition service is not needed base64 Coded.
A. TRUE
B. FALSE
Answer: B
255. Only matrix A and matrix B have the same number of rows and columns. A and B can be added.
A. True
B. False
Answer: A
256. Which of the following is true about unsupervised learning?
A. Unsupervised algorithm only processes "features" and does not process ·tags·
B. Dimensionality reduction algorithm is not unsupervised learning
C. K-means algorithm and SVM algorithm belong lo unsupervised learning
Answer: A
257. Which of the following options is not a tensor attribute?
A. name (Logo)
B. shape (Dimensionality)
C. type (Type)
D. node (Node)
Answer: D
258. What are the characteristics of Python code? (Multiple Choice)
A. leg1bil1ty
B. Simpl1c1ty
67
C. Rapidity
D. Scalab1hty
Answer: AB D
259. As shown in the figure below, what is the value of the determinant A?
1 2 3
A- 4 5 6
7 8
A. 24
B. 18
C. -24
D. 0
Answer: D
260. What is the most important difference between batch gradient descent, mini-batch gradient descent,
and stochastic gradient descent?
A. Gradient size
B. Gradient direction
C. Learning rate
D. Number of samples used
Answer: D
261. Which of the following does not belong to automatic hyperparameter optimization algorithm?
A. Grid search
B. Random gradient descent
C. Random search
D. Model-based hyper parameter optimization
Answer: B
68
262. Which of the following conditions do the randomized trials need to meet? (Multiple Choice)
A. It can be repeated under the same conditions.
B. There may be more than 011e possible outcome for each trial. and all possible outcomes of the trial can
be clarified m advance
C. All results of the lest cannot be clarified in advance
D. It Is not possible to determine which result will appear before conducting a test.
Answer: AB D
263. Which description is wrong about the lambda function of the Python language?
A. lambda is just an expression, and the function body is much simpler than def.
B. The body of a lambda can be an expression or a block of code
C. The lambda function can access the parameters in the global namespace.
D. The lambda function accepts only one parameter value
Answer: D
264. The face detection service will report an error when there is no face in the test image.
A. TRUE
B. FALSE
Answer: B
265. The current technology application directions of artificial intelligence mainly include?
B. Control System
C. Computer vision
D. Speech Recognition
Answer: AC D
266. The following evaluation indicators belonging to the regression algorithm are?
A. Recall rate
69
B. Confusion matrix
C. Mean square error
D. Accuracy
Answer: C
267. In the process of deep learning model training, what are the common optimizers?
A. Adam
B. Adagrad
C. SGD
D. Momentum
Answer: A B C D
268. GPU Good at computationally intensive and easy to parallel programs.
A. TRUE
B. FALSE
Answer: A
269. All convolution kernels of the same convolution layer in the convolutional neural network are
weight-sharing.
A. TRUE
B. FALSE
Answer: A
270. The loss function reflects the error between the target output and actual output of the neural network.
The commonly used loss function in deep learning is:
A. Exponential loss function
B. Mean square loss function
C. Log loss function
D. Hinge Loss function

70
Answer: B
271. Huawei Cloud El builds enterprise intelligence services based on three-tier services. Which of the
following options does not belong to Layer 3 services?
A. Basic platform services
B. General domain services
C. Industry sector services
D. Integration services
Answer: D
272. When training the network of God, we often encounter many problems. For the problem of vanishing
gradient, we can choose to make. Which of the following functions to alleviate the problem?
A. Relu function
B. Sigmoid function
C. tanh function
D. Softsign function
Answer: A
273. MSIR It is a simple, efficient and flexible graph-based function IR, A functional semantic type that can
be represented Have?
A. Free variable
B. Higher order function
C. Interrupt
D. Recursion
Answer: AB D
274. Which of the following options is not a reason for traditional machine learning algorithms to promote
the development of deep learning?
A. Dimensional disaster
71
B. local invariance and smooth regularization
C. Manifold learning
Answer: D
275. TensorFlow2.0 not support tensor in GPU with CPU Transfer between each other.
A. TRUE
B. FALSE
Answer: B
276. Which of the following is not MindSpore common Operation?
A. signal
B. math
C. nn
D. array
Answer: A
277. Among the machine learning algorithms, the following is not an integrated learning strategy?
A. Boosting
B. Stacking
C. Bagging
D. Marking
Answer: D
278. Under the large-scale relationship analysis scenario, which of the following options does not belong to
the three high demands of massive relationship processing?
A. Efficient relationship discovery data for massive data
B. Efficient sharing of massive data
C. Efficient storage and access requirements for massive amounts of data

72
D. High scalability and high availability requirements for relational analysis platforms
Answer: B
279. What are the commonly used gradient descent optimization functions? (Multiple Choice)
B. Adadelta
C. Adagrad
D. momentum
E. RMSProp
Answer: A B C D E
280. Code model.fit(mnist.train.inmage,mnis t.train.labels.epochs=5)in of epochs Parameter
representative?
A. The entire training set will be trained 5 Times
B. The entire test set will be tested 5 Times
C. The entire training set will be divided into 6 Share
D. The entire training set will be divided into 5 Share
Answer: A
281. Which of the following is not part of the deep learning development framework?
A. CNTK
B. Keras
C. SAFA
D. MXNet
Answer: C
282. TensorF1ow Which of the following companies first developed?
A. Oracle
B. Facebook
73
C. Nvidia
D. Google
Answer: D
283. Which of the following steps in deep learning are automatically completed by the model?
A. Model training
C. Analysis and positioning task
D. Feature extraction
Answer: B D
284. HUAWEI HiAI Empower APP What value?
A. Safety
B. stable
C. real time
D. at any time
Answer: A B C D
285. HUAWEI HiAI Which module does the face detection in the platform belong to?
A. HiAI Engine
B. HiAI Framework
C. HiAI Foundation
D. HiAI Service
Answer: A
286. From the labeled historical data, we can predict that the retail sales of the next quarter will be in 20-30
Wan still 30-40 Ten thousand", What is the problem?
A. Regression problem
B. Rule problem
74
C. Classification problem
D. Clustering problem
Answer: C
287. Label when the image remake detection service is called successfully sugestion What's the situation?
A. certainty
B. FALSE
C. uncertainty
D. TRUE
Answer: BCD
288. Which of the following aspects belong to Huawei's full stack Al solution?
A. Ascend
B. CANN
C. ModelArts
D. MindSpore
Answer: A B C D
289. Huawei firmly believes that the value of Al is ultimately reflected in solving the actual business
problems of the enterprise. Therefore, Huawei is firmly committed to the development of Al technology,
focusing on domain solutions, focusing on basic technologies and enabling platforms, and working with
industry practice leaders to develop industry solutions
A. True
B. False
Answer: A
290. What are the results returned by the if conditional statements in the Python language? (Multiple
choice)
A. True
75
B.O
C. FALSE
D. null
Answer: BC
291. What is wrong description or backpropagation?
A. The learning process or the backpropagation algorithm consists of a forward propagation process and a
back-propagation process.
B. The backpropagation algorithm is a learning algorithm suitable for multi-layer neural networks, which is
based on the gradient descent method
C. The backpropagation phase sends training inputs to the network to obtain an stimuli response
D. until the response of the network to the input reaches the predetermined target range.
Answer: C
292. Huawei Ascend Al Chip is NPU (Neural network processor) One of the typical representatives.
A. TRUE
B. FALSE
Answer: A
293. Ascend among the following options 310 And Shengteng 910 The attributes are the same?
A. Maximum power
B. Architecture
C. Integer precision
D. 7nm Process
Answer: B
294. What are the implementation modes ofTensorflow? (Multiple Choice)
A. Stand-alone mode
B. D1stnbuted mode
76
C. Reverse mode
D. Forward mode
Answer: AB
295. According to Huawei Cloud El Intelligent platform, which of the following solutions can be provided?
A. Crowd statistics program
B. Policy query scheme based on knowledge graph
C. Vehicle identification scheme
D. Intrusion recognition scheme
Answer: A B C D
296. Bagging In integrated learning, the relationship between each base learner is?
A. Additive relationship
B. relationship
C. The latter model must be built on top of the previous model
D. Independent
Answer: B
297. Which of the following description 1s wrong about the image recognition cloud service?
A. The service identifies specific objects such as skyscrapers, buildings , and ships.
B. The service can also identify the sea, city, port and other scenes.
C. The service can perform semantic analysis on the content in the image to extract labels with specific
abstract meanings
D. The service can provide object and scene labels and can also provide more conceptual labels.
Answer: D
298. On the premise of ensuring data privacy and security, federated learning utilizes different data sources
to cooperatively train models to improve breakthroughs in data bottlenecks.
A. TRUE
77
B. FALSE
Answer: A
299. Which of the following services belong to Huawei El Service family?
A. Dialogue robot
B. El Basic services
C. Natural language processing
Answer: A B C D
300. When the voice recognition service is successfully called, which field is the recognition result stored
in?
A. result
B. content
C. data
D. text
Answer: A
301. About the image content review service returned when the call is successful suggestion Field, the
correct statement is?
A. pass Representative does not contain sensitive information, passed
B. review The representative needs a manual review
C. suggestion The field represents whether the test passed
D. block Representative contains sensitive information and does not pass
Answer: A B C D
302. Which of the following is not the difference between Python 2 and Python 3?
A. print
B. Unicode
78
C. import
D. xrange
Answer: C
303. Which of the following is not an application of image recognition services?
A. Target Detection
B. Smart photo album
C. Scene analysis
D. Speech synthesis
Answer: D
304. Which of the following statements about Python are correct? (Multiple choice)
A. Invented m 1989 by the Dutch Guido van Rossum, the first public release was issued in 1991.
B. Python 1s purely free software and the source code follows the GPL (GNU General Public License)
protocol
C. Python syntax is simple and clean One of the features is to force blank characters to be used as
statement 1ndentat1on [Right Answers)
D. Python 1s often nicknamed glue language which can easily connect various modules made 1n other
languages
Answer: A B C D
305. Python authors deliberately design very restrictive grammars that make bad programming habits (such
as the next line of if statement not indented to the right) cannot be compiled
A. True
B. False
Answer: A
306. It is fine to pass in only one image when calling the face comparison service.
A. TRUE
79
B. FALSE
Answer: B
307. HiAI Mobile computing platform provides rich upper-level functional services API, Can run efficiently
on mobile devices.
A. TRUE
B. FALSE
Answer: A
308. What are the common clustering algorithms?
A. Density clustering
B. Hierarchical clustering
C. Spectral clustering
D. Kmeans
Answer: A B C D
309. X, Y are random variables, C is a constant, the nature of the difference in the following options which is
wrong? (Multiple choice)
A. D(C)=O
B. D(X+ Y)=D(X)+D(Y)
C. D(CX)=C'C'D()()
D. D(XY)=D(X)D{Y)
Answer: B D
310. According to the definition of information entropy what is the bit entropy of throwing a uniform coin?
A. 0
B. 0.5
C. 1
D. -1
80
Answer: C
311. In machine learning, what input the model needs to train itself and predict the unknown?
A. Manual procedure
B. Neural Networks
C. Training algorithm
D. historical data
Answer: D
312. When the universal text recognition service is successfully called, the returned text recognition result is
stored in which of the following fields?
A. text
B. result
C. content
D. words
Answer: B
313. Which of the following description about the relationship between the Poisson distribution and the
binomial distribution is correct?
A. The mathematical model of the Poisson distribution and the binomial distribution are both Lagrangian
models.
B. Poisson d1stnbutJon 1s an approx1mallon of the bmom1al d1stnbubon when n 1s very large and p 1s
small (
C. Poisson distribution has no relationship with binomial distribution
D. Poisson distribution can replace binomial distribution
Answer: B
314. GAN is a deep learning model and one of the most promising methods for unsupervised learning in
complex distribution in recent years.

81
A. True
B. False
Answer: A
315. The activation function plays an important role in the neural network model learning and
understanding of very complex problems. The following statement about the activation function is correct.
A. Activation functions are linear functions
B. Activation functions are non-linear functions
C. The activation function is partly a nonlinear function, partly a linear function
D. Most of the activation functions are nonlinear functions, and a few are linear functions
Answer: B
316. Feedforward neural network is a simple neural network, each neuron is arranged hierarchically, it is
currently one of the most widely used and fastest growing artificial neural networks. The following statement
about feedforward neural networks is correct:
A. Neurons with computing power are connected to the upper and lower layers
B. Its input node has computing power
C. Connecting neurons in the same layer
D. Information only along-Pass through
Answer: D
317. The determinant of square matrix A is a scalar.
A. True
B. False
Answer: A
318. The Python language does not allow embedding another loop body in a loop body
A. True
B. False
82
Answer: B
319. The following does not belong TensorFlow2.0 Is characterized by?
A. Multi-core CPU accelerate
B. distributed
C. multi-language
D. Multi-platform
Answer: A
320. The word recognition in the speech recognition service refers to the synchronous recognition of short
speech. Upload the entire audio at once, and the recognition result will be returned in the response.
A. TRUE
B. FALSE
Answer: A
321. About Python correct tuple, which is correct description? (Multiple choice)
A. The difference between a list and a tuple is that the elements of the tuple cannot be
B. The list uses square brackets and the tuple uses parentheses
C. Tuple creation is very simple just add elements in parentheses and separate them wrth commas
D. When a tuple contains only one element you need to add a comma after the element
Answer: A B C D
322. . enter 32*32 Image with size 5*5 The step size of the convolution kernel is 1 Convolution
calculation, output image Size is:
A. 28*23
B. 28*28
C. 29*29
D. 23*23
Answer: B
83
323. The optimizer is an important part of training neural networks. The purpose of using the optimizer does
not include which of the following:
A. Speed up algorithm convergence
B. Reduce the difficulty of manual parameter setting
C. Avoid overfitting
D. Avoid local extremes
Answer: C
324. What operating systems does Python support? (Multiple Choice)
A. DOS
B. Linux
C. Mac OSX
D. Windows
Answer: A B C D
325. Corvolutional neural networks are more suitable for dealing with speech recognition problems
A. True
B. False
Answer: B
326. What does factors that promote the development of artificial intelligence not include?
A. Big data
B. Computing ability
C. Algorithm theory
D. Block chain
Answer: D
327. Deep learning makes it easy to derive simple mathematical functions from a large amount of
84
high-dimensional data to describe complex decision interfaces.
A. True
B. False
Answer: B
328. Which of the following is the computing model of TensorFlow?
A. Calculation chart
B. Tensor
C. Conversation
D. Variable
Answer: A
329. What is the subject of artificial intelligence?
A. Mathematics and Physiology
B. Psychology and Physiology
C. Linguistics
D. Comprehensive mterd1sc1pl1nary and marginal d1sc1phnes
Answer: D
330. Add to the loss function of linear regression L1 Regular term, this time the regression is called Lasso
return.
A. TRUE
B. FALSE
Answer: A
331. Image label service returned tag There can be more than one.
A. TRUE
B. FALSE
Answer: A
85
332. TensorF1ow2.0 Can be used to check if it is tensor The methods are?
A. dtype
B. isinstance
C. is_tensor
D. device
Answer: BC
333. Regarding the face search service, which of the following statements are correct?
A. When there is no face set, you need to create a face set first, then add face data, and then search
B. The size of a face set cannot exceed 10000 Pictures
C. There is a dedicated interface to delete the specified face set
D. There is a dedicated interface to delete the face data in a certain face set
Answer: AC D
334. The commonly used loss functions in deep learning are?
A. L1 Loss function
B. Mean square error loss function
C. Cross entropy error loss function
D. Self-declining loss function
Answer: BC
335. In May 1997, the famous "Human-Machine Wars" final computer defeated Kasparov, the world chess
king, with a total score of 3.5 to 2 5. is this computer called?
A. Dark blue
B. Dark green
C. Ponder
D. Blue sky
Answer: A
86
336. HUAWEI HiAI Which tool does the platform support to integrate with?
A. Jupyter Notebook
B. MyEclipse
C. Android Studio
D. Spider
Answer: C
337. Artificial intelligence is a new technical science that studies and develops theories, methods and
application systems for simulating, extending and extending human intelligence. It is one of the
core research areas of machine learning.
A. True
B. False
Answer: B
338. X. Y are random variables. C is a constant. which of the following description about the nature of the
mathematical expectation is incorrect?
A. E(C) = CA. E(C) = C

B. E(X+Y) = E(X)+E{Y)
C. E(CX) = CE(X)
D. E(XY) = E(X)E{Y)
Answer: D
339. The continue statement in the Python language is used to jump out of this loop and then continue to
execute the next round of statements.
A. True
B. False
Answer: A
87
340. Which of the following libraries are installed Atlas300 (3000)Install the accelerator card in the server
DDK (Device Development Kit) What needs to be installed before?
A. decorator
B. setuptools
C. numpy
D. tensorflow
Answer: ABC
341. Python regular expressions are a special sequence of characters that makes it easy to check if a string
matches a pattern.
A. True
B. False
Answer: A
342. The matrix produced by the exchange of matrix A rows and columns is called the transpose of A .
What are the correct properties of matrix transposition? (Multiple Choice)
A. (AT) T = A
B. (A+ 8) T = AT +BT
C. (!EA) T = !EAT
D. (AB} T = AT +BT
Answer: ABC
343. Which of the following are Al Application areas?
A. Wisdom education
B. Smart City
C. Smart home
D. Smart medical
Answer: A B C D
88
344. Principal Component Analysis (PCA) is a statistical method. A set of variables that may be related to
each other is transformed into a set of linearly related variables by orthogonal transformation
The converted set of variables is called the principal component
A. True
B. False
Answer: B
345. The following about KNN Algorithm k The value description is correct?
A. K The larger the value, the easier the model is to overfit
B. K The larger the value, the smoother the segmentation surface of the classification
C. K Value is a hyperparameter
D. can k Value is set to 0
Answer: BC
346. HUAWEI HiAI What are the supported models?
A. P30
B. Mate 20
C. glory V20
D. iPhone10
Answer: ABC
347. Atlas 800 Al There are multiple models of servers, of which the one based on the Kunpeng
processor platform is?
A. Atlas 800 model: 9000
B. Atlas 800 model: 3000
C. Atlas 800 model: 3010
Answer: B
348. Self-encoder is an unsupervised learning algorithm

89
A. True
B. False
Answer: A
349. In the process of training the neural network, we use the gradient descent method to continuously
update which value, which makes the loss Function minimization?
A. Number of samples
B. Eigenvalues
C. Hyperparameter
D. parameter
Answer: D
350. The following statement about recurrent neural networks is wrong?
A. Recurrent neural network can be unfolded according to the time axis
B. LSTM Unable to solve the problem of vanishing gradient
C. LSTM It is also a recurrent neural network
D. Recurrent neural network can be abbreviated as RNN
Answer: B
351. According to the American Society of Automotive Engineers (SAE) The automatic driving is divided
into Which level?
A. L1~L4
B. L1~L5
C. L0~L4
D. L0~L5
Answer: D
352. It is known that the total parameter of a certain layer of the fully connected neural network is 330,The
number of neurons in the previous layer and this layer May be?
90
A. 32 with 10
B. 10 with 33
C. 33 with 10
D. 9 with 33
Answer: BC
353. Which of the following features does Python support? (Multiple choice)
A. Inheritance
B. Heavy load
C. Derived
D. Multiple inheritance
Answer: A B C D
354. Training error will reduce the accuracy of the model and produce under-fitting. How to improve the
model fit? (Multiple choice)
A. Increase the amount of data
B. Feature Engineering
C. Reduce regularization parameters
D. Add features
Answer: AB D
355. When training the network, we often encounter many problems. For the problem of gradient
disappearance, we can alleviate the problem by choosing which of the following functions to use?
A. Softsign function
B. Relu function
C. tanh function
D. Sigmoid function
Answer: B
91
356. Numerical calculation refers to the method and process of effectively using the digital computer to
solve the approximate problem of mathematical problems, and the discipline consisting of related theories.
Which of the following processes are involved in solving actual problems with a computer? (Multiple
Choice)
A. Know the actual problem
B. Mathematical model
C. Numerical calculation method
D. Programm1ng
E. Computer calculation results
Answer: A B C D E
357. What is wrong description of the Python module?
A. The Python module is a Python file that ends with .py and contains Python object definitions and Python
statements.
B. The Python module allows you to logically organize your Python code snippets.
C. Python modules can define functions classes and variables but the module does not contain
executable code
D. Assigning relevant code to a module can make your code better and easier to understand.
Answer: C
358. TensorFlow is Google's first generation of proprietary machine learning systems
A. True
B. False
Answer: B
359. Which of the following functions can numerically stabilize overflow and underflow?
A. Softminus function
B. Softplus function
C. Soft max function

92
D. Softmin func!Jon
Answer: C
360. Which of the following options are <Artificial Intelligence Comprehensive Experiment> May be used in
calling Huawei cloud services?
A. ak
B. region
C. project_id
D.sk
Answer: A B C D
361. The perceptron can be displayed in space as?
A. line
B. flat
C. Hyperplane
D. point
Answer: C
362. In neural networks, which of the following methods are used to update the parameters when training
the network to minimize the loss function?
A. Forward propagation algorithm
B. Pooling calculation
C. Convolution calculation
D. Backpropagation algorithm
Answer: D
363. The timestamps in the Python language- are represented by how long (in seconds) elapsed from
midnight (epoch) on January 1, 1970
A. True
93
B. False
Answer: A
364. For the image classification problem, which of the following neural networks is more suitable to solve
this problem?
A. sensor
C. Convolutional Neural Network
D. Fully connected neural network
Answer: C
365. Python is a fully object-oriented language. Which of the following options belong to the Python object?
(Multiple Choice)
A. Function
B. Module
C. Number
D. Character string
Answer: A B C D
366. View Atlas300 (3000) Which command should be used if the accelerator card driver is installed
successfully?
A. npu-sim info
B. npu info
C. atlas-driver info
D. atlas info
Answer: A
367. Which of the following can optimization problem be classified according to the constraints? (Multiple
choice)
94
A. equality constraints
B. Inequality constraints
C. Unconstrained conditions
D. Semi-constraint conditions
Answer: ABC
368. Huawei Al The full scenarios include public cloud, private cloud, various edge computing , loT industry
terminals, and consumer terminals and other end, edge, and cloud deployment environments.
A. TRUE
B. FALSE
Answer: A
369. Which of the following products is related to artificial intelligence? (Multiple Choice)
A. Alpha Go
B. Self-driving
C. Voice input
D. Huawei Mate mobile phone
Answer: A B C D
95
1. Faced with the challenge of achieving efficient distributed training for ultra-large-scale models, MindSpore is handled as?
A. Automatic parallel
B. Serial
C. Manual parallel
Answer: A
2. Which of the following is not the difference between Python 2 and Python 3?
A. print
B. Unicode
C. import
D. xrange
Answer: C
3. Which of the following options is not central to linear algebra?

A. Probability theory
B. Linear transformation
C. Matrix theory
D. Vector space
Answer: A
6. Huawei Machine learming Service MLS MLS is a one-stop platform that supports the entire process of data analysis.
Which of the following is not a feature of MLS?
A. A rich library of machine learning algorithms.
B. machine learning program is intuitive and easy to use.
C. Distributed and scalable big data computing engine.
D. Support for the R language but does not support the Python language
Answer: D
7. GBDT Compared with the random forest algorithm, which of the following statements is wrong?
A. GBDT Algorithm is easier to underfit than random forest
B. Random forest is calculated in parallel, while GBDT Can't
C. GBDT Algorithm is easier to overfit than random forest
D. GBDT And random forest are built on CART Based on the tree
Answer: C
8. The following code was used when compiling the model: model. compile(optimizer='Adam, loss='categorical.
crossentropy', metrics=[tf. keras. metrics. accuracy]). currently using evaluate
When the method evaluates the model, which of the following indicators will be output?
A. accuracy
B. categorical_1oss
C. loss
D. categorical accuracy
Answer: AC
9. According to Huawei Cloud El Intelligent platform, which of the following solutions can be provided?
A. Crowd statistics program
B. Policy query scheme based on knowledge graph
C. Vehicle identification scheme
D. Intrusion recognition scheme
Answer: ABCD
10. Which of the following does not belong to long-term memory LSTM(long Short Term Memory)network architecture?
A. Memory door
B. Forget the door
96
C. Input door
D. Output door
Answer: A
11. Functions are well-organized, non-reusable code segments used to implement a single, or associated Function.
A. True
B. False
Answer: B
12. Which of the following options is not the Tensorflow build process?
A. Building a calculation chart
B. Input tensor
C. Generate a session
D. Update weights I Right Answers)
Answer: D
13. Artificial intelligence is the research and development of theories, methods and application systems used to simulate,
extend and expand human intelligence Of a new technological science.
A. TRUE
B. FALSE
Answer: A
14. Tensor Flow Which of the following companies first developed?
A. Oracle
B. Facebook
C. Nvidia
D. Google
Answer: D
15. In Huawei Cloud El Enterprise Intelligence, which basic platform services are included?(Multiple Choice)
A. Machine learning
B. Deep leaming
C. Graph engine
D. Batch processing
Answer: A B CD
16. The activation function plays an important role in the neural network model leaming and understanding of very complex
problems. The following statement about the activation function is correct.
A. Activation functions are linear functions
B. Activation functions are non-linear functions
C. The activation function is partly a nonlinear function, partly a linear function
D. Most of the activation functions are nonlinear functions, and a few are linear functions
Answer: B
17. PyTorch Which of the following functions does not have?

A. Inline keras
B. Support dynamic graph
C. Automatic derivative
D. GPU accelerate
Answer: A
18. What is wrong description of the Python module?

A. The Python module is a Python file that ends with py and contains Python object definitions and Python statements.
B. The Python module allows you to logically organize your Python code snippets.
C. Python modules can define functions classes and variables but the module does not contain executable code
D. Assigning relevant code to a module can make your code better and easier to understand.
Answer: C
19. What are the commonly used activation functions?(Multiple Choice)

A. sigmoid
97
B. tanh
C. relu
D. danish
Answer: ABC
20. What are the Python language data types?(Multiple Choice)
A. numbers
B. string
C. list
D. tuple
E. dictionary
Answer: ABCDE
21. When the voice recognition service is successfully called, which field is the recognition result stored in?
A. result
B. content
C. data
D. text
Answer: A
22. What does factors that promote the development of artificial intelligence not include?
A. Big data
B. Computing ability
C. Algorithm theory
D. Block chain
Answer: D
23. The image data of the passport recognition service is not needed base64 Coded.
A. TRUE
B. FALSE
Answer: B
24. The loss function of logistic regression is the cross-entropy loss function.
A. TRUE
B. FALSE
Answer: B
26. It is fine to pass in only one image when calling the face comparison service.
A. TRUE
B. FALSE
Answer: B
27. Which of the following options is not the-session mode used by Tensorflow?
A. Explicitly call the session to generate function
B. Explicitly call the session to close function
C. Through the Python context manager
D. Multiple POST queries
Answer: D
28. twenty two Among machine leaming algorithms, which of the following is not unsupervised learming?
A. GMM
B. Xgboost
C. Clustering
D. Association rules
Answer: B
29. Which of the following is not included in the recurrent neural network usage scenario?
A. Machine translation
B. Speech recognition
C. Image style migration
98
D. Text generation
Answer: C
30. What are the characteristics of Python code?(Multiple Choice)

A. legibility
B. Simplicity
C. Rapidity
D. Scalability
Answer: ABD
31. In a convolutional neural network, different layers have different functions. Which of the following layers can play the
role of dimensionality reduction?
A. Input layer
B. Fully connected layer
D. Pooling layer
Answer: BCD
32. After the data has completed the feature engineering operation, in the process of constructing the model, which of the
following options is not a step in the decision tree construction process?
A. Pruning
C. Data cleaning
D. Decision tree generation
Answer: C
33. Which method is not supported in TensorFlow to define variables?

A. Random number
B. Constant
C. Calculated from the initial values of other variables
D. null
Answer: D
34. Which of the following descriptions about the Recurrent Neural Network(RNN)is correct?
A. Can be used to process sequence data.
B. Cannot process variable length sequence data.
C. Unlike convolutional neural networks. parameters of RNN cannot be shared
D. The units above the hidden layer are not associated With each other.
Answer: A
35. In order for a machine to be intelligent, it must be knowledgeable. Therefore, there is a research field in artificial
intelligence. which mainly studies how computers automatically acquire knowledge and skills to achieve self-improvement
What is the branch of this research called?
A. Expert system
B. Machine learning
C. Neural Network
Answer: B
36. TensorFlow only supports Python and R language

A. True
B. False
Answer: B
37. The naive Bayes algorithm does not require independent and identical distribution among sample features.
A. TRUE
B. FALSE
Answer: B
99
38. What are the scenarios or industries that are suitable for using Python?(Multiple Choice)
A. Artlficial intelligence
B. web development
C. Game development
D Hardware development
Answer: ABC
39. Which is wrong with the Python module time introduction method?
A. import time
B. from time import*
C. import time as t
D. from time
Answer: D
40. Huawei's chip support HUAWEI HiAl Which module of?

A. HiAl Engine
B. HiAl Foundation
C. HiAl Framework
D. HiAl Service
Answer: B
41. There are a lot of data generated during the training of the neural network. What mechanism does TensorFlow use to
avoid excessive input data?
A. Client
B. feed
C. placeholder
D. fetch
Answer: C
42. The optimizer is an important part of training neural networks. The purpose of using the optimizer does not include
which of the following:
A. Speed up algorithm convergence
B. Reduce the difficulty of manual parameter setting
C. Avoid overfitting
D. Avoid local extremes
Answer: C
43. The TensorFlow framework does not support Windows systems.

A. True
B. False
Answer: B
44. The following is the correct difference between machine learning algorithms and traditional rule-based methods?
A. Traditional rule-based methods, in which the rules can be explicitly clarified manually
B. Traditional rule-based methods use explicit programming to solve problems
C. The mapping relationship of the model in machine learning is automatically leamed
D. The mapping relationship of the model in the machine learning institute must be implicit
Answer: ABC
45. In machine leaming, what input the model needs to train itself and predict the unknown?
A. Manual procedure
B. Neural Networks
C. Training algorithm
D. historical data
Answer: D
46. Traffic agent can achieve()Hour full time/Global traffic awareness?

A. 7x24
B. 5x24
100
C. 5x12
D. 7x12
Answer: A
47. In the process of training the neural network, our goal is to keep the loss function reduced. Which of the following
methods do we usually use to minimize the loss function?
A. Gradient descent
B. Dropout
C. Cross-validation
D. Regularization
Answer: A
48. Deflated Al The logical architecture of the processor does not include which of the following options?
A. DVPP
B. GPU
C. Al Calculation engine
D. Chip system control CPU
Answer: B
51. Numerical calculation refers to the method and process of effectively using the digital computer to solve the
approximate problem of mathematical problems, and the discipline consisting of related theories.
Which of the following processes are involved in solving actual problems with a computer?(Multiple Choice)
A. Know the actual problem
B. Mathematical model
C. Numerical calculation method
D. Programming
E. Computer calculation results
Answer: ABCDE
52. What people say about neural networks is wrong?

A. As the number of hidden layers of the neural network increases, the classification ability of the model gradually weakens
B. The limitation of a single-layer perceptron is that it cannot solve the XOR problem
C. The feedforward neural network can be represented by a directed acyclic graph
D. There is no connection between neurons in the same layer of the feedforward neural network
Answer: A
53. Which of the following conditions is not a condition that n-fold bernoulli trials needs to meet?
A. Each test was repeated under the same conditions.
B. There are only two possible outcomes for each trial, ie. event A occurs and event A does not occur.
C. Each test itself is obeying normal distribution
D. The results of each trial are independent of each other.
Answer: C
54. In May 1997, the famous"Human-Machine Wars"final computer defeated Kasparov, the world chess king, with a total
score of 3. 5 to 2 5. is this computer called? \
A. Dark blue
B. Dark green
C. Ponder
D. Blue sky
Answer. A
55. The number or matrix is stored in the tensor

A. True
B. False
Answer: B
57. Machine Leaming Service is a data mining and analytics platform service that helps users quickly discover data pattens
and build predictive models through machine learning techniques and deploy them as predictive analytics solutions
A. True
101
B. False
Answer: A
58. Which of the following does the convolutional neural network structure not include?
B. Pooling layer
C. Loop layer
D. Fully connected layer
Answer: C
59. If a model has a large deviation on the test set and a small variance, it means that the model?
A. Overfitting
B. May be overfitting may be underfitting
C. Just fit
D. Underfitting
Answer: D
60. In the face search service, if we want to delete a certain face set, we can use this code: firs_cliet. get_v2().
get_face_set_service(). delete_face_set("""), among them **"" is to fill in the actual face set name.
A. TRUE
B. FALSE
Answer: A
61. Jobs that are repetitive and require weak social skills are the easiest to be Al Replaced work.
A. TRUE
B. FALSE
Answer: A
62. Regular term can also be added to logistic regression to avoid overfitting.
A. TRUE
B. FALSE
Answer: A
64. The continue statement in the Python language is used to jump out of this loop and then continue to execute the next
round of statements.
A. True
B. False
Answer: A
65. enter 32*32 Image with size 5*5 The step size of the convolution kernel is 1 Convolution calculation, output image Size
is:
A. 28*23
B. 28*28
C. 29*29
D. 23*23
Answer: B
66. Which of the following is not MindSpore common Operation?

A. signal
B. math
C. nn
D. array
Answer: A
67. The current technology application directions of artificial intelligence mainly include?
B. Control System
C. Computer vision
D. Speech Recognition
102
Answer: ACD
68. In TensorFlow, data is represented in the form of tensors and calculation charts.
A. True
B. False
Answer: B
69. What are the conditions for m row n column matrix A and p row q column matrix B to be multiplied?
A. m=p. n=q
B. n=p
C. m=n
D. p=q
Answer: B
70. The following about the general form identification service returned type The field statement is correct?
A. type Representative text recognition area type
B. type for text Time represents the text recognition area
C. type Representative form type
D. type for table Time represents the form recognition area
Answer: BD
71. Which of the following are the characteristics of the Python language?{Multiple choice)
A. Explanatory
B. Process oriented
C. Object-oriented
D. Dynamic data type
Answer: ACD
75. What does not belong to supervised leaming?

B. Support vector machine
C. Decision tree
D. Principal component analysis
Answer: D
77. Tensorflow is the second generation of artificial intelligence learning system developed by Google based on ().
A. DistBelief
B. PaleyFunction
C. ConvexOne
D. Infinity
Answer: A
78. When training the network, we often encounter many problems. For the problem of gradient disappearance, we can
alleviate the problem by choosing which of the following functions to use?
A. Softsign function
B. Relu function
C. tanh function
D. Sigmoid function
Answer: B
79. Which is the correct description of the python creation function?(Multiple choice)
A. The function created starts with the def keyword followed by the function name and parentheses.
B. The parameters need to be placed in parentheses
C. The function content starts with a colon and needs lo be indented (Right Answers)
D. Return the result with return and the function ends.
Answer: AB CD
80. Recurrent neural networks can capture dynamic information in serialized data.
A. TRUE
B. FALSE
Answer: A
103
81. Al Chips are divided into business applications and can be divided into?
A. training
B. GPU
C. Model building
D. reasoning
Answer: AD
83. Which of the following is not a module in the Tensorflow library?

A. tf. nn
B. tf. layers
C. tf. boost
D. tf. contrib
Answer: C
84. Corvolutional neural networks are more suitable for dealing with speech recognition problems
A. True
B. False
Answer: B
85. Which of the following activation functions are prone to vanishing gradient problems?
A. ReLU
B. Softplus
C. Tanh
D. Sigmoid
Answer: AC
86. Which of the following products is related to artificial intelligence?(Multiple Choice)

A. Alpha Go
B. Self-driving
C. Voice input
D. Huawei Mate mobile phone
Answer: ABCD
87. In the process of deep learning model training, what are the common optimizers?
A. Adam
B. Adagrad
C. SGD
D. Momentum
Answer: ABCD
89. Huawei Ascend Al Chip is NPU (Neural network processor)One of the typical representatives.
A. TRUE
B. FALSE
Answer: A
90. Linear algebra is a branch of algebra that deals with linear problems. The linear relationship is that the relationship
between exponential objects is expressed in ( )times form?
A. 1
B. 2
C. 3
D. 4
Answer: A
91. The model composed of machine learning algorithms cannot represent the true data distribution function on a
theoretical level. Just approach it.
A. TRUE
B. FALSE
Answer: A
104
92. X, Y are random variables, C is a constant, the nature of the difference in the following options which is wrong?
(Multiple choice)
A. D(C)=0
B. D(X+Y)=D(X)+D(Y)
C. D(CX)=CCD00
D. D(XY)=D(X)D(Y)
Answer: BD
93. Which of the following does not belong to automatic hyperparameter optimization algorithm?
A. Grid search
B. Random gradient descent
C. Random search
D. Model-based hyper parameter optimization
Answer: B
95. Python is a fully object-oriented language. Which of the following options belong to the Python object?(Multiple Choice)
A. Function
B. Module
C. Number
D. Character string
Answer: ABCD
96. The following options do not belong to the scope of Huawei's full-stack solution are?
A. Application enable
B. Edge computing
C. Open source framework
D. Chip enable
Answer: B
97. In which year was the first introduction of "artificial intelligence"?

A. 1946
B. 1960
C. 1916
D. 1956
Answer: D
98. Feature is the dimension that describes the characteristics of the sample.
Regarding its interpretability in traditional machine learning and deep learming, the following statement is correct
A. Features are interpretable in traditional machine learning, but weak in deep leaming
B. Features are weak in traditional machine learning, but strong in deep leaming
C. Features are weak in interpretability in traditional machine leaming and deep learning
D. Features are interpretable in traditional machine leaming and deep learming
Answer: A
101. Artificial intelligence is a new technical science that studies and develops theories, methods and application systems
for simulating, extending and extending human intelligence. It is one of the core research areas of machine learning.
A. True
B. False
Answer: B
102. The loss function reflects the error between the target output and actual output of the neural network. The commonly
used loss function in deep learning is:
A. Exponential loss function
B. Mean square loss function
C. Log loss function
D. Hinge Loss function
Answer: B
103. Which of the following is an image label service URI?

105
A. /v1.0/image/recognition
B. /v1.0/image/celebrity-recognition
C. /v1.0/image/recapture-detect
D. /v1.0/image/tagging
Answer: D
104. Convolutional neural networks are more suitable for image recognition problems than cyclic neural networks.
A. True
B. False
Answer: A
105. In neural networks, weights are often share D.

Which of the following neural networks will share weights?
A. sensor
D. Recurrent neural network
Answer: B and C
106. Python authors deliberately design very restrictive grammars that make bad programming habits (such as the next line
of if statement not indented to the right) cannot be compiled
A. True
B. False
Answer: A
107. Which of the following items are included in the results returned when the face search service is successfully called?
A. Searched face similarity
B. Searched faces id
C. Searched face position
D. Searched face number
Answer: A, B and C
108. Tf. keras. Data sets can view keras the built-in data set.
A. TRUE
B. FALSE
Answer A
109. TensorFlow2.0 of keras preprocessing the role is?

A. keras Data processing tools
B. keras Built-in optimizer
C. keras Model deployment tool
D. Keras Generative model tool
Answer: A
110. HiAl Mobile computing platform provides rich upper-level functional services API, Can run efficiently on mobile
devices.
A. TRUE
B. FALSE
Answer: A
111. In a neural network, knowing the weight and deviations of each neuron is the most important step. If you know the
exact weights and deviations of neurons in some way, you can approximate any function. What is the best way to achieve
this?
A. Random assignment, pray that they are correct
B. Search for a combination of weight and deviation until the best value 1s obtained
C. Assign an initial value to iteratively update weight by checking the difference between the best value and the initial
D. The above is not correct
Answer: C
112. What are the mainstream deep learning open source tools? (Multiple choice)
106
A. Tensor Flow
B. Caffe
C. Torch
D. Theano
E. Scikit Jeam
Answer: A, B, C and D
113. Which of the following options belong to Tensor flow concept? (Multiple Choice)
A. Tensor
B. Variables
C. Placeholder
D. Operation
E. Conversation
Answer: A, B, C, D and E
114. The following evaluation indicators belonging to the regression algorithm are?
A. Recall rate
B. Confusion matrix
C. Mean square error
D. Accuracy
Answer: C
115. The Python list can be identified by Il', and the default index of the first element from left to right is 1.
A. True
B. False
Answer: B
116. What is wrong description of the normal distribution?

A. In natural phenomena and social phenomena, many random variables obey or approximate a normal distribution.
B. The normal distribution lakes the maximum value at the mean
C. The larger the standard deviation the steeper the normal distribution curve
D. The larger the standard deviation, the slower the normal distribution curve.
Answer: C
117. The Python dictionary is identified by "{}", and the internal data consists of the key and its corresponding value.
A. True
B. False
Answer: A
118. Which of the following are solutions for the Huawei Cloud El industry scenario?(Multiple Choice)
A. Intelligent Logistics
B. Intelligent Water
C. Intelligent transportation
D. Intelligent Finance
E. Intelligent manufacturing
121. Which of the following options is not a tensor attribute?

A. name (Logo)
B. shape (Dimensionality)
C .type (Type)
D. node (Node)
Answer: D
122. On the premise of ensuring data privacy and security, federated learning utilizes different data sources to
cooperatively train models to improve breakthroughs in data bottlenecks.
A. TRUE
B. FALSE
Answer: A
107
123. The following description of machine learning is correct?

A. Deep learning is a branch of machine learning
B. Deep learning and machine learning are mutually inclusive
Deep learning and machine learning are both artificial intelligence but have no relationship to each other
Answer: A
124. The following about the standard RNN Model, the correct statement is?
A. There is no one-to-one model structure
B. Do not consider the time direction when back propagating
C. There is no many-to-many model structure
D. There will be a problem of attenuation of long-term transmission and memory information
Answer: D
125. None of the following options TensorFlow2.0.The supported operators are:

A. pow
B. @
C. ^
D. //
Answer: C
127. Which of the following steps in deep learning are automatically completed by the model?
A. Model training
C. Analysis and positioning task
D. Feature extraction
Answer: B and D
128. About the image content review service returned when the call is successful suggestion Field, the correct statement
is?
A. pass Representative does not contain sensitive information, passed
B. review the representative needs a manual review
C. suggestion the field represents whether the test passed
D. block Representative contains sensitive information and does not pass
129. Which of the following are the elements of artificial intelligence?(Multiple choice)
A. Algorithm
B. Computing power
C. Data
D. Scene
130. In the environment preparation for invoking HUAWEI CLOUD services, which of the following does not require
confirmation of installation?
A. requests
B. setuptools
C. websocket-client
D. Jieba
Answer: D
131. Which of the following are the activation functions of deep learning algorithms?
A. Sigmoid
B. ReLU
C. Tanh
D. Sin
Answer: A, B and C
133. ModelArts Service and ().The combination of services can easily deploy the model to "end"?
A. OBS
108
B. OCR
C. ECS
D. HiLens
Answer: D
134. Which of the following is not an artificial intelligence school?

A. Symbolism
B. StatJst1caltsm
C. Behaviorism
D. Connectionism
Answer: B
135. What is not the optimization method in deep learning?

B. Back propagation algorithm
C. Principal component analysis
D. Momentum
Answer: C
136. GAN is a deep learning model and one of the most promising methods for unsupervised learning in complex
distribution in recent years.
A. True
B. False
Answer: A
137. Which of the following is true about unsupervised learning?
A. Unsupervised algorithm only processes "features" and does not process tags
B. Dimensionality reduction algorithm is not unsupervised learning
C. K-means algorithm and SVM algorithm belong lo unsupervised learning
Answer: A
140. Which of the following is not Mind Spore Features of the core architecture?
A. Automatic differentiation
B. Automatic tuning
C. Automatic coding
D. Automatic parallel
Answer: C
141. Atlas 800 Al There are multiple models of servers, of which the one based on the Kunene processor platform is?
A. Atlas 800 model: 9000
B. Atlas 800 model: 3000
C. Atlas 800 model: 3010
Answer: B
142. As the following, What are the Python language design philosophy? (Multiple Choice)
A .Beautiful
B. Expensive
C. Explicit
D. Simple
Answer: A and D
143. Tensor flow Operations and Computation Graph are not-run in the Session
A. True
B. False
Answer: B
144. The meaning of artificial intelligence was first proposed by a scientist in 1950, and at the same time a test model of
machine intelligence was proposed who is this scientist?
A. Minsky
109
B. Zade
C. Turing
D. Von Neumann
Answer: C
147. The for loop statement in the Python language can iterate through the items in any sequence.
A. True
B. False
Answer: A
148. What are the algorithms supported by Tensorflow?(Multiple Choice)

A. GNN
B. ZNN
C. RNN(Right Answers)
D.HUT
Answer: A and C
149. Which of the following is the computing model of Tensor Flow?

A. Calculation chart
B. Tensor
C. Conversation
D. Variable
Answer: A
150. HUAWEI HiAI Empower APP what value?

A. Safety
B. stable
C. real time
D. at any time
151. When we describe the house, we usually use the attributes such as residential area, house type, decoration type, etc.
If we use plain With Bayes as a model, we assume that there is no relationship between attributes.
A. TRUE
B. FALSE
Answer: A
153. Huawei Al The full scenarios include public cloud, private cloud, various edge computing, loT industry terminals and
consumer terminals and other end, edge and cloud deployment environments.
A. TRUE
B. FALSE
Answer: A
154. Which of the following options are<Artificial Intelligence Comprehensive Experiment>May be used in calling Huawei
cloud services?
A. ak
B. region
C. project_id
D. sk
155. Tensor Flow2.0 not support tensor in GPU with CPU Transfer between each other.
A. TRUE
B. FALSE
Answer: B
156. About Bayesian formula-P(WIX)=P(XIW)'P(W)/P(X)What is the correct description?

A. P(WIX) is a prior probability
B. P(XIW) 1s a conditional probability
110
C. P(W) is the posterior probability

D. P(X) is the posterior probability
Answer: B
157. Tensor Flow2.0 of keras.metrics the methods that are not included in the built-in indicators are?
A. Accuracy
B. Recall
C. Sum
D. Mean
Answer: C
159. What are the implementation modes of Tensor flow?(Multiple Choice)

A. Stand-alone mode
B. D1stnbuted mode
C. Reverse mode
D. Forward mode
Answer: A and B
160. Use with Atlas300 (3000) What conditions need to be checked when the accelerator card server compiles and runs
the program?
A. Carry out Atlas Driver Installation
B. It has been installed CUDA Software package
C. It has been installed CMAKE Compilation tool
D. Complete environment configuration
Answer: A, C and D
161. Only matrix A and matrix B have the same number of rows and columns .A and B can be added.
A True
B. False
Answer: A
162. The training error will continue to decrease as the model complexity increases.
A. TRUE
B. FALSE
Answer: A
163. Which of the following services belong to Huawei El Service family?

A. Dialogue robot
B. El Basic services
C. Natural language processing
164. What is wrong description or back propagation?

A. The learning process or the back propagation algorithm consists of a forward propagation process and a back-
propagation process.
B. The back propagation algorithm is a learning algorithm suitable for multi-layer neural networks, which is based on the
gradient descent method
C. The back propagation phase sends training inputs to the network to obtain an stimuli response
D. Until the response of the network to the input reaches the predetermined target range.
Answer: C
166. Which of the following description about Python regular expression is not correct?
A. Python has added there regular expression module since version 1.5.
B.re regular expression module gives the Python language all the regular expression functionality.
C.re regular expressions are powerful tools for working with strings.
D.re regular expressions can only process string data not numeric data
Answer: D
167. The number of rows and columns of the three matrices A, B. and C is 3 rows and 2 columns, 2 rows. 3 columns.3
111
rows and 3 columns which of the following operations is meaningful?

A.AC
B.BC
C.A+B
D.AB-BC
Answer: B
168. Which of the following description 1s wrong about the image recognition cloud service?
A. The service identifies specific objects such as skyscrapers, buildings, and ships.
B. The service can also identify the sea, city, port and other scenes.
C. The service can perform semantic analysis on the content in the image to extract labels with specific abstract meanings
D. The service can provide object and scene labels and can also provide more conceptual labels.
Answer: D
169. In random forest, what strategy is used to determine the outcome of the final ensemble model?
A. Cumulative system
B. Find the average
C. Voting system
D. Cumulative system
Answer: B and C
170. Al Chip is also called Al Accelerators are specially designed to handle a large number of computing tasks in artificial
intelligence applications functional module.
A. TRUE
B. FALSE
Answer: A
175. About Python correct tuple, which is correct description? (Multiple choice)
A. The difference between a list and a tuple is that the elements of the tuple cannot be
B. The list uses square brackets and the tuple uses parentheses
C. Tuple creation is very simple just add elements in parentheses and separate them worth commas
D. When a tuple contains only one element you need to add a comma after the element
Answer: A,B,C,D
176. Atlas 200 DK Biting 310 the main application of the chip is?
A. Model reasoning
B. Build model
C. Training model
Answer: A
177.With Atlas300(3000)Install the accelerator card in the server DDK(Device Development Kit)Only need to install when
dependent on the environment Python3.
A.TRUE
B.FALSE
Answer: A
178. In TensorFlow2.0 in ft. contra the method can continue to be used.

A.TRUE
B.FALSE
Answer: B
179. What are the regularizations in deep learning? (Multiple Choice)

A.L1 norm.L2 norm
B .Data set enhancement
C. Integration method
D. Dropout
Answer: A, B, C, D
180. Which of the following statements about universal form recognition services are correct?
A. Arrows Represents the line information occupied by the text block, the number is from 0 Start, list form
112
B. columns Represents the column information occupied by the text block, the number is from 0 Start, list form
C. The incoming image data needs to go through base64 coding
D. words Representative text block recognition result
Answer: A, B, C, D
181. The timestamps in the Python language-are represented by how long (in seconds) elapsed from midnight (epoch) on
January 1, 1970
A. True
B. False
Answer: A
182. In a neural network based on connectionism, each node can express a specific meaning.
A.TRUE
B.FALSE
Answer: B
183. The Python language can use multiple statements on the same line separated by commas
A. True
B .False
Answer: B
184. Huawei Al The whole scenario includes public cloud, private cloud various edge computing lot industry terminals and
consumer .The deployment environment of terminal, edge ,and cloud for fee-based terminals
A.TRUE
B.FALSE
Answer: A
186. What of the following does belong to convolutional neural network (CNN)? (Multiple Choice)
A. VGG Net
B. Res Net
C. Alex Net
D. Google Net
Answer: ABCD
187. The word recognition in the speech recognition service refers to the synchronous recognition of short speech Upload
the entire audio at once and the recognition result will be returned in the response.
A.TRUE
B.FALSE
Answer: A
188. In what year did Huawei officially provide services in the form of cloud services, and combined with more partners to
provide richer artificial intelligence practices?
A.2002
B.2013
C.2015
D.2017
Answer: D
189. In the gradient descent algorithm, which of the following algorithms is the most confusing algorithm for the trajectory
on the loss function surface?
A.SGD
B.BGD
C.MGD
D.MBGD
Answer: A
190. Which of the following are included in the application scenarios of Ensor flow? (Multiple choice)
A. Speech recognition
B. Face recognition
C. Image style changes
113
D. Autopilot
Answer: A, B, C, D
192. An e-commerce company has always had problems with official website stalls, poor utilization of network resources
and difficulty in operation and maintenance. Then which of the following can the company use El service?
A. Park Agent
B. Traffic Agent
C. Network agent
D. Industrial Agent
Answer: C
193. Regarding the convolutional neural network pooling layer, the following description is correct?
A. Pooling operation is realized by scanning window
B. The pooling layer can play a role in dimensionality reduction
C. Commonly used pooling methods include maximum pooling and average pooling
D. The pooled feature image becomes smaller
Answer: ABCD
194. If the deep learning neural network has the problem of gradient disappearance or gradient explosion, our common
solution is.
A. Gradient shear
B. Random under sampling
C. use Rely Activation function
D. Regularization
Answer: ACD
195. Python script execution mode includes interactive mode and script mode
A. True
B. False
Answer: A
196. Principal Component Analysis (PCA)is a statistical method .A set of variables that may be related to each other is
transformed into a set of linearity related vandals by orthogonal transformation The converted set of variables is called the
principal component
A. True
B. False
Answer: B
198. Training error will reduce the accuracy of the model and produce under-fitting. How to improve the model fit? (Multiple
choice)
A. Increase the amount of data
B. Feature Engineering
C. Reduce regularization parameters
D. Add features
Answer: ABD
199. Which of the following is not a way for the Tensor Flow program to read data?
A. Preload data
B .Feeding data
C. Read from the file
D. Write a file format reader
Answer: D
200. The following statement about recurrent neural networks is wrong?
A. Recurrent neural network can be unfolded according to the time axis
B. LSTM Unable to solve the problem of vanishing gradient
C. LSTM It is also a recurrent neural network
D. Recurrent neural network can be abbreviated as RNN
Answer: B
201. What is the service categories included in Huawei Cloud El Enterprise Intelligence? (Multiple Choice)
A. El visual cognition
B. El online games
C. El speech semantics
D .El industry scene
Answer: A, C D
114
202. Which of the following description of the validation set is wrong?

A. The verification set can coincide with the test set.
B. The test set can coincide with the training set
C. The subset used to pick hyper parameters is called a validation set
D. Typically 80% of the training data 1s used for training and 20% 1s used for verification.
Answer: A
204. In the deep learning network the back propagation algorithm is used to find the optimal parameters.
What rules are used in the back propagation algorithm to obtain the derivation layer by layer?
A. Chain rule
B. Cumulative rule
C. Law of reciprocity
D. Rule of Normalization
Answer: A
205 Artificial intelligence is currently suitable for scenarios with I' known environment, clear objectives, and predictable
action ±Deep learning in the fields of image recognition, speech recognition, translation,
Etc. artificial intelligence basically has human recognition ability,even surpassing human beings Based on these
capabilities.it has been applied to many scenes such as medical care and public safety
However, there is still a lack of reasoning and cognition
A. True
B. False
Answer: A
206. Which of the following neural network structures will share weights? (Multiple choice)
A. Convolutional neural network
D. All of the above
Answer: A and B
207. Which of the following about the description of expectations and variances is and variance are both numerical
characteristics of random variables
D. The greater the expectation the smaller the variance
Answer: D
209. In deep learning, if the neural network has a large number of layers, the problem of gradient disappearance is more
likely to occur.
Strictly speaking, in which of the following links does the sample disappearance problem occur?
A. Back propagation update parameters
B. Forward the dream and update
C .Back propagation calculation result
D. Forward calculation results
Answer: A
210. Among the following properties TensorFlow2.0 Does not support creation tensor .The method is?
A. zeros
B. fills
C. create
D. constant
Answer: C
211. Which of the following are the cloud services provided by Huawei Cloud El Visual Cognition?(Multiple choice)
A. Text recognition
B. Face recognition
C. Image recognition
D. Content detection
E. Image processing
Answer: A, B, C, D, E
115
212. Which of the following conditions do the randomized trials need to meet? (Multiple Choice)
A. It can be repeated under the same conditions.
B. There may be more than 011e possible outcome for each trial, and all possible outcomes of the trial can be clarified m
advance
C. All results of the lest cannot be clarified in advance
D. It Is not possible to determine which result will appear before conducting a test.
Answer: A, B, D
213. HUAWEI HiAI Which module does the face detection in the platform belong to?
A. HiAl Engine
B. HiAl Framework
C. HiAl Foundation
D. HiAl Service
Answer: A
215. The Python language does not allow embedding another loop body in a loop body
A. True
B. False
Answer: B
216.Tensor Flow2.0 The methods that can be used for tensor merging are?
A. join
B. concert
C. split
D. unstick
Answer: B
217. What are the commonly used gradient descent optimization functions? (Multiple Choice)
B. Ad delta
C. Anagram
D. momentum
E. RMS Prop
Answer: A, B, C, D, E
218. Which of the following statement about the matrix is incorrect?

A. The arbitrary matrix is multiplied by the unit matrix and will not change.
B. The transpose of the symmetric matrix A 1s also A itself.
C. The transpose of the orthogonal matrix A is equal to the inverse of A.
D. There is no inverse matrix for the diagonal matrix.
Answer: D
219.Bagging In integrated learning, the relationship between each base leaner is?
A. Additive relationship
B. relationship
C. The latter model must be built on top of the previous model
D. Independent
Answer: B
221. MSIR It is a simple, efficient and flexible graph-based function IR,A functional semantic type that can be represented
Have?
A. Free variable
C. Interrupt
D. Recursion
Answer: A,B and D
222. Linear regression in 3 The fitted surface in the dimension above dimension is?
A. Curved surface
116
B. flat
C. Hyper plane
D. Hyper surface
Answer: C
223. Which of the following features does Python support?(Multiple choice)

A. Inheritance
B. Heavy load
C. Derived
D. Multiple inheritance
224. Tensor flow supports multi-TPU cluster computing

A. True
B. False
Answer: A
225. What quotation marks can the Python language use?(Multiple Choice)
A. Single quotes
B. Double quotes
C. Three quotes
D. Four quotes
Answer: A,B and C
227. Which of the following options is not a reason for traditional machine learning algorithms to promote the development
of deep learning?
A. Dimensional disaster
B. local invariance and smooth regularization
C. Manifold learning
Answer: D
228. From the perspective of technical architecture Al The classification of chips includes?
A.FPGA
B.CPU
C.GPU
D.ASIC
229. X.Y are random variables. C is a constant. which of the following description about the nature of the mathematical
expectation is incorrect?
A. E(C) = CA.E(C) = C
B. E(X+Y) = E(X)+E{Y)
C. E(CX)=CE(X)
D. E(XY)=E(X)E{Y)
Answer: D
230. Traditional machine learning and deep learning are the core technologies of artificial intelligence. There is a slight
difference in the engineering process. The following steps.
What you don't need to do in deep learning is:
A. Model evaluation
C. Data cleaning
D. Model building
Answer: B
231. Which of the following descriptions about Python program are correct? (Multiple choice)
A. The program consists of modules
B. Module contains statements
117
C. statement contains an expression

D. Expressions create and process objects
Answer: A,B,C and D
233. The correlation coefficient. also known as the linear correlation coefficient, Is used to measure the
linear relationship between two variables, which is a real number greater than zero
A. True
B. False
Answer: B
A.TRUE
B.FALSE
Answer: B
235. HUAWEI HiAl what are the supported models?

A.P30
B. Mate 20
C. glory V20
D.iPhone10
Answer: A, B and C
236. Which description is wrong about the hyper parameter?

A.1-typerparameters are parameters that set values before the algorithm begins learning
B. Most machine learning algorithms have hyper parameters.
C. Hyper parameters cannot be modified
D. The value of the hyper parameter is not lamed by the algorithm itself.
Answer: C
237. In supervised learning, "Those who are near Zhu are red and those who are near Mo are black "is used to describe
which of the following models?
A.K-Means
B.SVM
C.KNN
D. Neural Networks
Answer: C
238. What are the commonly used loss functions? (Multiple choice)
A. Mean variance
B. Sigmoid cross entropy
C. Soft max cross enl10py
D. Sparse cross entropy
E. Weighted S1gmo1d Cross Entropy
239. There are many commercial applications for machine learning services. What are the main business scenarios
covered?(Multiple Choice)
A. Financial product recommendation
B. Predictive maintenance
C. Telecom customer retention
D. Retailer grouping
240. Root The user cannot install Atlas300 (3000) Install the accelerator card on the server DDK (Device Development
Kit),Can only be installed by ordinary users.
A.TRUE
B.FALSE
Answer: A
118
241. The test error will keep getting smaller as the complexity of the model increases.
A. TRUE
B. FALSE
Answer: B
242. Voice recognition refers to the recognition of audio data as text data.
A. TRUE
B. FALSE
Answer: A
243. The pooling layer in the convolutional neural network can reduce the size of the lower layer input. Common pooling is:
A. Minimum strata
B. Product pooling layer
C. Maximum pooling layer
D. Average pooling layer
Answer: C D
244. Deep learning makes it easy to derive simple mathematical functions from a large amount of High-dimensional data to
describe complex decision interfaces.
A. True
B. False
Answer: B
246. Tensor Flow in keras The built-in optimizer of the module has?
A. Adam
B. SGD
C. Adaboost
D. Adadelta
Answer: A B C D
247: Artificial intelligence at this stage is still in the weak artificial intelligence stage.
A. TRUE
B. FALSE
Answer: A
248. Global gradient descent algorithm, stochastic gradient descent algorithm and batch gradient descent algorithm are all
gradient descent algorithms. The following is wrong about its advantages and disadvantages
A. The global gradient algorithm can find the minimum value of the loss function
B. Batch gradient algorithm can solve the local minimum problem
C. Stochastic gradient algorithm can find the minimum value of the loss function
D. The convergence process of the global gradient algorithm is time-consuming
Answer: C
119
249. In the deep learning neural network,the perceptron is the simplest neural network. The correct statement about its
structure is:
A. There are only two hidden layers
B. Only one hidden layer
C. Its network uses Sigmoid Activation function
D. Its network uses Relu Activation function
Answer: B
250. Which of the following data types does Tensorflow not support?
A. int8
B. ftoat32
C. double64
D. bool
Answer: C
252. K Folding cross-validation refers to dividing the test data set into K Sub-data sets.
A. TRUE
B. FALSE
Answer: B
253. It is known that the total parameter of a certain layer of the fully connected neural network is 330,The number of
neurons in the previous layer and this layer May be?
A. 32 with 10
B. 10 with 33
C. 33 with 10
D. 9 with 33
Answer: B C
254. Add to the loss function of linear regression L1 Regular term,this time the regression is called Lasso return
A. TRUE
B. FALSE
Answer: A
255. Which of the following description about the relationship between the Poisson distribution and the binomial distribution
is correct?
A. The mathematical model of the Poisson distribution and the binomial distribution are both Lagrangian models.
B. Poisson d1stnbutJon 1s an approx1mallon of the bmom1al d1stnbubon when n 1s very large and p 1s small (
C. Poisson distribution has no relationship with binomial distribution
D. Poisson distribution can replace binomial distribution
Answer: B
256. Ce11 Provides basic modules for defining and performing calculations, Ce11 The object can be
executed directly, the following statement is wrong?
A. __init__, Initialization parameters(Parameter), Submodule(Ce11), operator(Primitive)Equal group Software for initial
verification
B. Construct, Define the execution process. In graph mode, it will be compiled into graphs for execution, and there is no
syntax restriction
C. There's some left optim Commonly used optimizers, wrap Pre-defined commonly used network packaging functions
Ce11
D. bprop (Optional),The reverse of the custom module
120
Answer: A C D
258. Which of the following schools does the neural network study belong to?
A. Symbolism
B. Connectionism
C. Behaviorism
Answer: B
259.What is wrong about the image content review service?

A. politics Test results for sensitive persons involved in politics
B. terrorism Test results for political violence
C. confidence Represents confidence, range 0-100
D. labe1 Label representing each test result
Answer: C
260. Which of the following can optimization problem be classified according to the constraints? (Multiple choice)
A. equality constraints
B. Inequality constraints
C. Unconstrained conditions
D. Semi-constraint conditions
Answer: A B C
261. Which of the following can improve the computational efficiency of the neural network model? (Multiple choice)
A.TPU
B.GPU
C.FPGA
D. Large-scale distributed cluster
Answer: A B C D
262. Feedforward neural network is a simple neural network,each neuron is arranged hierarchically, it is
currently one of the most widely used and fastest growing artificial neural networks.
The following statement about feedforward neural networks is correct
A. Neurons with computing power are connected to the upper and lower layers
B. Its input node has computing power
C. Connecting neurons in the same layer
D. Information only along-Pass through
Answer: D
263. L1 with L2 Regularization is a method commonly used in traditional machine learning to reduce
generalization errors. The following is about the two.
The right way is:
A. L1 Regularization can do feature selection
B. L1 with L2 Regularization can be used for feature selection
C. L2 Regularization can do feature selection
D. L1 with L2 Regularization cannot be used for feature selection
Answer: A
264. Which description is wrong about the lambda function of the Python language?
A. lambda is just an expression, and the function body is much simpler than def.
B. The body of a lambda can be an expression or a block of code
121
C. The lambda function can access the parameters in the global namespace.
D. The lambda function accepts only one parameter value
Answer: D
265. How many spaces does PEP 8 stipulate that Python's first fine indentation needs to indent?
A. 1
B. 2
C. 4
D. 8
Answer: C
266. Which of the following environments does not support to install the T11nsorFlow?
A. Linux
B. Mac OS
C. Docker
D. OpenStack
Answer: D
267. Ascend among the following options 310 And Shengteng 910 The attributes are the same?
A. Maximum power
B. Architecture
D. 7nm Process
Answer: B
268. The time interval in the Python language is a Hoating-point fraction in seconds
A. True
B. False
Answer: A
270. Reducing the gap between the training error and the test error will result in over-fitting How to prevent over-fitting?
(Multiple Choice)
A. Cross validation
B. Integration method
C. Increase regularization
Answer: A B C D
271. When dealing with actual problems, when should machine learning be used in the following situations?
A. The data distribution itself changes over time, requiring continuous re-adaptation of the program, such as predicting the
trend of merchandise sales
B. The complexity of the rules is low, and the problem is small
C. Task rules will change over time, such as defect detection on the production line
D. The rules are very complicated or cannot be described,such as face recognition and voice recognition
Answer: A C D
272. In neural networks, which of the following methods are used to update the parameters when training the network to
minimize the loss function?
B. Pooling calculation
C. Convolution calculation
D. Backpropagation algorithm
122
Answer: D
273. What are the results returned by the if conditional statements in the Python language?(Multiple choice)
A. True
B. 0
C. FALSE
D. null
Answer: B C
275. TensorFlow 2.0 The way to view tensor dimensions in is?

A. dimens
B. dtype
C. ndim
D. device
Answer: C
276. When the universal text recognition service is successfully called, the returned text recognition result is stored in which
of the following fields?
A. text
B. result
C. content
D. words
Answer: B
277. Grid search is a method of parameter adjustment

A.TRUE
B.FALSE
Answer: B
278.As shown in the figure below,what is the value of the determinant A?

2 3
A. 24
B. 18
C. -24
D. 0
Answer: D
279. During the two classification process, we can set any category as a positive example.
A.TRUE
B.FALSE
Answer: A
282. Atlas accelerate Al What processor is used for inference?

A. Different 910 processor
B. Different 310 processor
123
C. GPU
D. FPGA
Answer: B
283. Label when the image remake detection service is called successfully sugestion .What's the situation?
A. certainty
B.FALSE
C. uncertainty
D. TRUE
Answer: B C D
284. HiAl What are the advantages of mobile computing platforms?

A. Quick source code
B. Diverse tool chain
C. Complete documentation
D. plentiful API
Answer: A B C D
285. The following applications that are not part of the adversarial generation network are?
A. Text generation
B. Image generation
C. Image identification
D. Data enhancement
Answer: C
286. What is the most important difference between batch gradient descent, mini-batch gradient descent, and stochastic
gradient descent?
A. Gradient size
C. Learning rate
D. Number of samples used
Answer: D
288. TensorFlow 2.0 stand by GPU The acceleration is mainly manifested in the parallel operation of the addition,
subtraction, multiplication and division of the matrix accelerate.
A.TRUE
B.FALSE
Answer: A
289. Generative confrontation networks are currently widely used. The following scenarios can use this network?
A. Data enhancement
C. Information retneval
D. Image generation
Answer: A B C D
295. Regarding the face search service, which of the following statements are correct?
A. When there is no face set, you need to create a face set first, then add face data, and then search
B. The size of a face set cannot exceed 10000 Pictures
C. There is a dedicated interface to delete the specified face set
D. There is a dedicated interface to delete the face data in a certain face set
124
Answer: A C D
296. Deep learning is a branch of machine learning

A. True
B. False
Answer: A
297. The main computing resources included in the Da Vinci architecture computing unit are?
A. Vector calculation unit
B. Scalar Computing Unit
C. Tensor computing unit
D. Matrix calculation unit
Answer: A B D
301. According to the definition of information entropy what is the bit entropy of throwing a uniform coin?
A. 0
B. 0.5
C. 1
D. -1
Answer: C
302. During application development, which of the following operations is not a typical network definition?
A. Network definition
B. Network execution
C. Knowledge distillation
D. Weight initialization
Answer: C
303. The determinant of square matrix A is a scalar.

A. True
B. False
Answer: A
306. Which of the following descriptions of the depth feedforward network is correct?
A. Deep feedforward network 1s a kind of neural network.
B. The deep feedforward network has only one hidden layer.
C. The unit on the hidden layer of the deep feedforward network will have countless.
D. Deep feedforward network 1s used to deal with linear problems.
Answer: A
307. The matrix produced by the exchange of matrix A rows and columns is called the transpose of A. What are the correct
properties of matrix transposition? (Multiple Choice)
A. (AT) T=A
B. (A+8) T=AT+BT
C. (IEA) T=IEAT
D. (AB) T=AT+BT
Answer: A B C
308. TensorFlow is Google's first generation of proprietary machine learning systems

A. True
B. False
Answer: B
125
310. The following options belong to keras under estimator The encapsulation methods are?
A. Assessment
B. training
C. prediction
D. Output model
Answer: A B C
311. With a lot of sales data but no labels, companies want to identify VIP Customer,the following model Suitable?
B. SVM
C. K-Means
D. Hierarchical clustering
Answer: C D
312. What operating systems does Python support? (Multiple Choice)
A. DOS
B. Linux
315. HUAWEI HiAI Which tool does the platform support to integrate with?
A. Jupyter Notebook
B. MyEclipse
C. Android Studio
D. Spider
Answer: C
316. The following belongs to TensorFlow2.0 The characteristic is?
A. Introduce Keras interface

B. Support static image mechanism
C. Support more platforms and more languages
D. Continue to be compatible Tensorflowl x Related modules
Answer: A C D
317. HUAWEI CLOUD ModelArts Is for Al Which of the following functions are in the developed one-stop development
platform Mode1Arts Can have?
A. Data govemance
B. Al market
C. Visual workflow
D. Automatic leaming
Answer: A B C D
318. Vector is a number.
A. True
B. False
Answer: B
319. Which of the following about the gradient descent is incorrect?
A. Random gradient descent is a commonly used one in gradient descent.

B. Gradient descent includes random gradient descent and batch gradient descent.
C. The gradient descent algorithm 1s fast and reliable [Right Answers)
D. Random gradient descent is one of the commonly used optimization algorithms in deep learning algorithms
126
Answer: C
320. TensorFlow It is one of the most popular deep learning frameworks.
A. TRUE
B. FALSE
Answer: A
321. The trace operation returns the sum of the diagonal elements of the matrix. Therefore, the trace of matrix A and its
transposed matrix are equal
A. True
B. False
Answer: A
322. Regularization is an important and effective technique for reducing generalization errors in traditional machine
learning. The following techniques are regular.
The technology is:
A. L1 Regularization
B. L2 Regularization
C. Dropout
Answer: A B C
323. What is the English abbreviation for Al?
A. Automatic Intelligence
B. Artificial Intelligence
C. Automatic Information
D. Artificial Information
Answer: B
324. According to the development process of the robot,it is usually divided into three generations, respectively are:
(Multiple Choice)
A. Teaching Reproduction Robot

B. Robot with sensation
C. Robots that will think
D. Intelligent robot
Answer: A B D
325. Python regular expressions are a special sequence of characters that makes it easy to check if a string matches a
pattern.
A. True
B. False
Answer: A
326. What research areas does artificial intelligence have? (Multiple choice)

B. Computer Vision
C. Machine learning
Answer: A B C D
327. Information theory is a branch of applied mathematics. The main research is to quantify how much information a
signal contains.
Who is the first to propose information theory?
127
A. Minsky
B. Turing
C. Simon
D. Shannon
Answer: D
328. On-Device Execution, that is, the entire image is offloaded and executed, and the computing power
Of the Yiteng chip can be fully utilized, which can greatly reduce the interaction overhead, thereby
Increasing the accelerator occupancy rate.
On-Device The following description is wrong?
A. Mind Spore Realize decentralized autonomy through adaptive graph optimization driven by gradient
data A11 Reduce, Gradient aggregation is in step, and calculation and communication are fully
Streamlined
B. Challenges of model execution under super chip computing power: Memory wall problems, high interaction overhead,
and difficulty in data supply. Partly in Host Executed, partly in Device Execution, interaction overhead is even much greater
than execution overhead, resulting in low accelerator occupancy
C. Mind Spore Through the chip-oriented depth map optimization technology, the synchronization wait is less, and the
"data computing communication" is maximized. The parallelism of "trust”, compared with training performance Host Side
view scheduling method is flat
D. The challenge of distributed gradient aggregation under super chip computing power ReslNet50 Single iteration 20ms
Time will be generated The synchronization overhead of heart control and the communication overhead of frequent
synchronization. Traditional methods require 3 Synchronization completed A11 Reduce, Data-driven method autonomy
A11 Reduce, No control overhead
Answer: C
330. Which of the following aspects belong to Huawei's full stack Al solution?
A. Ascend
B. CANN
C. Model Arts
D. Mind Spore
Answer: A B C D
331. Code model. Fit (mnist.train.inmage,mnist.train.labels.epochs=5)in of epochs Parameter representative?
A. The entire training set will be trained 5 Times

B. The entire test set will be tested 5 Times
C. The entire training set will be divided into 6 Share
D. The entire training set will be divided into 5 Share
Answer: A
332. HUAWEI CLOUD EI There are many types of agents according to different industries and application
Scenarios. Currently, Huawei Cloud El The agent has?
A. Industrial Agent
B. Traffic Agent
C. Park Agent
D. Car intelligence
Answer: A B C D
333. Which of the following libraries are installed Atlas300 (3000) Install the accelerator card in the server DDK (Device
Development Kit) What needs to be installed before?
A. Decorator
B. Setup tools
C. Numpy
D. Tensor flow
Answer: A B C
128
336. Which of the following are Al Subfield?
A. Machine learning
B. Computer vision
Answer: A B CD
337. TensorFlow2.0 The mechanism of graphs and conversations has been cancelled in.
A.TRUE
B.FALSE
Answer: B
338. TensorFlow It is an end-to-end open source platform for machine learning and deep learning.
A.TRUE
B.FALSE
Answer: A
339. Which of the following are Al Application areas?
A. Wisdom education
B. Smart City
C. Smart home
D. Smart medical
Answer: A B C D
340. All convolution kemels of the same convolution layer in the convolutional neural network are weight-sharing.
A. TRUE
B.FALSE
Answer: A
341. Which of the following statements about supervised learning is correct?
A. Decision tree is a supervised learning

B. Supervised learning cannot use cross-validation for training
C. Supervised learning is a rule-based algorithm
D. Supervised learning can be trained without labels
Answer: A
343. Which of the following are the topics of speech processing research?
A. Speech processing
B. Voiceprint recognition
D. Wake up
Answer: B C D
A.TRUE
B. FALSE
Answer: B
345. As shown below, which are the following matrix A characteristic

Value? (Multiple choice)
A= [3 3]
129
A. 2
B. -2
C. 4
D. 4
Answer: A D
346. Which descriptions are correct about python's index? (Multiple Choice)
A. Index from left to right defaults from 0

B. Index from left to right defaults from 1
C. Index from right lo left defaults from-1
D. Index from right to left defaults from 0
Answer: A C
347. The three stages of artificial intelligence include computational intelligence, (), Cognitive intelligence.
A. Weak artificial intelligence

C. Behavioral intelligence
D. Strong artificial intelligence
Answer: B
349. Which of the following statements about Python are correct? (Multiple choice)
A. Invented m 1989 by the Dutch Guido van Rossum, the first public release was issued in 1991.
B. Python 1s purely free software and the source code follows the GPL (GNU General Public License) protocol
C. Python syntax is simple and clean One of the features is to force blank characters to be used as statement
1ndentation[Right Answers)
D. Python 1s often nicknamed glue language which can easily connect various modules made 1n other languages
Answer: A B C D
350. Which of the following capabilities does the genetic knowledge map have?
A. Auxiliary case diagnosis

B. Disease prediction and diagnosis
C. Gene test report generation
D. Entity query
Answer: A C D
351. The commonly used loss functions in deep learning are?
A. L1 Loss function
B. Mean square error loss function
C. Cross entropy error loss function
D. Self-declining loss function
Answer: B C
352. When training the network of God, we often encounter many problems. For the problem of vanishing gradient, we can
choose to make.
Which of the following functions to alleviate the problem?
A. Relu function
B. Sigmoid function
C. tanh function
D. Softsign function
Answer: A
353. There are many types of neural networks in deep learning. The following neural network information
Is one-way propagation
130
A. LSTM
C. Recurrent neural network
D. GRU
Answer: B
355. Which of the following statements about overflow and underflow in numerical calculations is correct? (Multiple choice)
A. Underflow occurs when the number close to zero ls rounded to zero

B. Overflow occurs when the number close to zero 1s rounded to zero
C. Overflow occurs when a large number of levels are approximated as 1P or-1P
D. Underflow occurs when a large number of levels are approximated as 1P or-1P
Answer: A C
356. For the image classification problem, which of the following neural networks is more suitable to solve this problem?
A. Sensor
C. Convolutional Neural Network
D. Fully connected neural network
Answer: C
357. Deep learning algorithms can be divided into supervised learning and unsupervised learning.
A. True
B. False
Answer: A
358. Under the large-scale relationship analysis scenario, which of the following options does not belong?
To the three high demands of massive relationship processing?
A. Efficient relationship discovery data for massive data

B. Efficient sharing of massive data
C. Efficient storage and access requirements for massive amounts of data
D. High scalability and high availability requirements for relational analysis platforms
Answer: B
359. The constituent elements of the Python identifier include: numbers, letters, and underscores.
A. True
B. False
Answer: A
361. A scalar k is multiplied by matrix A equal to k and each of the numbers in A is multiplied.
A. True
B. False
Answer: A
362. The label predicted by the regression algorithm is
A. Self-deformation
B. Discrete
C. Strain type
D. Continuous
Answer: D
363. Huawei firmly believes that the value of Al is ultimately reflected in solving the actual business problems of the
131
enterprise. Therefore, Huawei is firmly committed to the development of Al technology. Focusing on domain solutions,
focusing on basic technologies and enabling platforms, and working with industry practice leaders to develop industry
solutions
A. True
B. False
Answer: A
364. On Huawei Cloud El which of the following can be AI Technology is integrated into the application
Scenarios of all walk of life Al the advantages of technology to improve efficiency and improve
Experience.
A. El Agent
B. OBS Object storage service
C. Cloud database
Answer: A
365. Which command can be checked Atlas 300(3000)Whether the accelerator card is in place?
A. 1spci|grep'npu'
B. 1spci|grep'd100'
C. 1spci grep'atlas'
D. Atlas info
Answer: B
366. HUAWEI HCAI Engine can easily combine multiple AI ability and app integrated.
A. TRUE
B. FALSE
Answer: A
367. What is the subject of artificial intelligence?
A. Mathematics and Physiology

B. Psychology and Physiology
C. Linguistics
D. Comprehensive mterd1sc1pilnary and marginal d1sc1phnes
Answer: D
368. Pytorch which company launched it first?
A. Baidu
B. Google
C. Facebook
D. Huawei
Answer: C
369. SVM what are the commonly used kernel functions in?
A. Gaussian kernel function

B. Polynomale kernel function
C. Sigmiod kernel function
D. Linear kernel
Answer: A B C D
370. What is in the category of artificial intelligence?
A. Action
132
B. Perception
C. Cognition
D. None of these
Answer: A B C
371. TensorFlow the data type used is?
A. Scalar
B. Vector
C. Tensor
D. Matrix
Answer: C
372. Tensorflow 2.0 middle keras the three main advantage of the interface and user-friendly, modular and combination,
easy to expand
A. TRUE
B. FALSE
Answer: A
373. Vector group a1=(1,1,1), a2=(0,2,5),a3=(1,3,6), which of the following options is correct?
A. Linear correlation
B. Linear independence
C. aHa2+a3=0
D. 2a1+a2+a3
Answer: A
374. Facing industry research and full scenarios AI .The huge gap between application, MindSpore Bridging the application
gap to help inclusiveness AI of technological innovation does not include which of the following?
A. New ways of collaboration

B. New programming language
C. New programming paradigm
D. New execution mode
Answer: B
375. Which of the following statements about passport recongnition services are correct?
A. Country code country code representing the country where the passport was issued
B. Nationality of representation holder
C. Passport _number representative passport number
D. Confidence information of related fields. The greater than confidence, the higher the reliability of the corresponding field
identified this time. In a statistical sense, the greater the confidence, the higher the accuracy
Answer: A B C D
376. What model is not a cyclic neural network?
A. RNN
B.LSTM
C. GBOT
D. GRU
Answer: C
377. What are application scenarios for the break statement in the Python language?
A. Any python sataement

133
B. While loop statement

C. foe loop statement
D. Nested loop statement
Answer: B C D
378. Self-encoder is an unsupervised learning algrothim?
A. TRUE
B. FALSE
Answer: A
379. Tensor yes MindSpore the storage component of the data
A.TRUE
B. FALSE
Answer: A
380. The face detection services will report an error when there is no face in the test image?
A. TRUE
B. FALSE
Answer: B
381. What are the numeric types of python?
A. Int (integer type)

B. Long (long integer type)
C. Float (floating point type)
D. Complex (complex number)
Answer: A B C D
382. Loss function and model function are the same thing?
A. TRUE
B. FALSE
Answer: B
383. Cyclic neural networks are suitable for dealing with image recognition problems
A. TRUE
B. FALSE
Answer: B
384. The commonly used functions for mathematical operation in python are basically in the math module and the cmath
module
A. TRUE
B. FALSE
Answer: A
385. Which is not a deep learning algorithm?
A. Ster-encoder
B. Convolutional neural networks
C. Recurrent neural networks
D. Support vector machine
Answer: D
134
1. The K-fold cross-validation is to divide a test dataset into K sub-datasets.

A. True
B. False
2. If the L1 regular expression is added to the loss function f linear regression, the regression is called Lass regression
A. True
B. False
3. Softmax cannot be used for binary classification.

A. True
B. False
4. Federated learning uses different data sources to train models, further breaking data bottle necks while ensuring data
privacy and security.
A. True
B. False
5. Computer vision is a science f how to make computers “see”.

A. True
B. False
6. In Keras, model.summary() is used t print network information.

A. True
B. False
7. The image module in TensorFlow 2.0 is used t operate images.

A. True
B. False
8. A recurrent neural network can capture dynamic information in serialized data.

A. True
B. False
9. An activation function is used to convert linear mapping into nonlinear mapping.

A. True
B. False
10. The training principle used by a recurrent neural network is called backpropagation through time (BPTT).
A. True
B. False
11. Attlas Al computing platform can be applied to different scenarios such as finance, power, manufacturing...
A. True
B. False
12. MindSpore can be installed in a CPU environment.

A. True
B. False
13. MindSpore is an independent and collaborative device-edge-cloud training and inference framework, it is user-friendly
in development and efficient in running.
A. True
B. False
14. The industrial Intelligent Twins uses beg data and Al technologies to provide a full series of services Converting
design, production logistics, sales, and service. It helps enterprises gain leading position.
A. True
B. False
135
15. The HUAWEI HiAl 3.0 platform supports development of non-Huawei mobile phones.
A. True
B. False
16. When the decision tree algorithm des regression, it usually uses () as the decision basis.
A. Gain
B. Gainratio
C. Gini
D. Standard deviation
17. In machine learning, which of the following is not a part of the process from obtaining data to officially putting the data
into a model?
A. Data standardization
B. Data cleansing
C. Data dimension deduction
D. Data visualization n
18. Which f the following is the fitting surface for linear regression in more than three dimensions?
A. Surface
B. Plane
C. Hyperplane
D. Hypersurface
19. Which f the following processor is used for inference f the Atlas 200DK Al accelerator card?
A. Ascend 910 Al processor
B. Ascend 310 Al processor
C. FPGA
D. GPU
20. Which f the following is not a feature of TensorFlow 2.0?

A. Multi-lingual
B. Multi-platform
C. Distributed
D. Multi-core CPU acceleration
21. Which f the following operators is not supported by TensorFlow 2.0?

A. //
B. @
C. pow
D. ^
22. What is the purpose of adding the dropout layer when creating a network layer?
A. Extracting image features
B. Functioning as a classifier
C. Preventing over fitting
D. Increasing the non-linearity
23. Which f the flowing functions is not provided by PyTorch?

A. GPU acceleration
B. Dynamic graphs
C. Embedded Keras
D. Automatic derivation
24. What is the function of keras.preprocessing of TensorFlow 2.0?

A. Keras data processing tool
B. Keras model generation tool
C. Keras model deployment tool
D. Keras built-in optimizer
136
25. Deep leaning neutral network training requires a large amount of matrix computation Generally, hardware K required
for a computer to be Capable of parallel computing, Which f the following hardware devices can provide the parallel
computing capability?
A. CPU
B. DIMM
C. GPU
D. Mainbo0ard
26. Which of the following is continuously updated using the gradient descent method during neural network training to
minimize a loss function?
A. Hyperparameter
B. feature value
C. Number of samples
D. Parameter
27. Which of the following statements about neural networks is incorrect?

A. The limitation of a single-layer perceptron is that it cannot resolve XR problems.
B. As hidden layers of a neural network increase, the model classification capability gradually weakens.
C. The neurons at the same layer of the feedforward neural network are not interconnected.
D. A feedforward neural network can be represented using a directed acyclic graph.
28. Which f the following statements about optimizers is correct?

A. The learning rate of the momentum optimizer does not need t be set manually.
B. The learning rate of the AdaGrad optimizer is not automatically updated.
C. The momentum optimizer is updated with the same learning rate for each iteration.
D. There is n relationship between the Adam optimizer and AdaGrad optimizer.
29. An objective of neural network training is to reduce a loss function. Which of the following methods is commonly used to
A. Dropout
B. Gradient descent
C. Normalizer
D. Cross verification
30. LI and L2 regularization are common methods used by traditional machine learning to reduce generalization errors.
Which of the following statements about Lt and 12
regularization is correct?
A. L2 regularization can be used for feature selection
B. L1 regularization can be used for feature selection
C. Both L1 and L2 regularization can be used for feature selection.
D. L1 and L2 regularization cannot be used for feature selection.
31. In a deep leaning network, the backpropagation algorithms is used to search for optimal parameters, when of the
following rules is used in the backpropagation
algorithms to perform layer-by-layer derivation?
A. Chain rule
B. Accumulation rule
C. Equality rule
D. Normalization
32. Dataset expansion can effectively affect verifitting ccurrence, Which f the fllawing statements abut dataset expansin is
crrect?
A. Alanger dataset indicates 4 lwer vertitting prbability.
B A larger dataset indkates a higher verfitting prbability,
C. Asmatler dataset indicates a lwer verfitting prbatality.
D. An enlarged r shrunk dataset indicates a lwer verfitting prbability,
33. There are many types f neural netwrks in deep learning. Which f the fllwing transmits infrmatin ne-way prpagatin?
137
A. Cnvlunteuiral network
B. Recurneurrael network
C.
D. GRU
34. Features are dimensins that describe sample characterntics. Which f the fllwing is 4 Crrect statement abut the
interpretabilityf features #1 traditinal machine learning and deep fearningy?
A. Fstures have high interpretability in bth traditinal machine teaming and deep teatning.
B. Features have lw interpretability in bth traditinal machine learning and deep learning
C. Features have high interpretabibty in tracitinal machine leaming but lw interpretability in deep jearreng,
D. Festures have lw interpretability in traditinal machine leaming but high interpeetatiity in deep learning
35. Hw many layers des the sftware stack f an Ascend Al chip cnsist f?
A. 3
B. 4
C. 2
D. 5
36. Which f the fllwing is the main applicatin f the Ascend 310 chip serving as the basis f the Atlas 200 DK?
A. Mdel training
B. Mdel building
C. Mdel inference
37. Cell prvides basic mdules fr defining and executing cmputing. Cell bjects can be directly executed, Which f the fllwing
statements is incrrect?
A. _init_ initializes and verifies cmpnents such as Parameter Ce, and Primitive.
B. construct defines the xecutpircesns. In graph mde, a graph cmpiled fr executin, withut syntax restrictin.
C. bprp ptinal indicates the reverse directin f custmized mdules.
D. There are ats sme predefined culls such as cmmn ptimizers fr ptim and cmmn netwrk packaging furxtins fr wrap.
38. Which f the fllwing technlgy innvatins is nt used by MindSpre t bridge the huge gap between industry research and ail-
scenari Al applicatins and fcilittate inetusive AS?
A. New prgramming paradige
B. New executin paradigm
C. New way f cperatin
D. New prgramming unguage
39. Which f the fllwing is nt the applicatin f the image recgnitin service?
A. Scenari analysis
B. Smart album
C. bject detectin
D. Speech synthesis
40. Which f the fllwing functins is prvided by the HUAWEI HiAl Fundatin mdule?
A. Push services based n user requirements.
B. Integrate apps.
C. Enable services t autmatically target and acquire users.
D. Quickly cnvert and prt existing mdels.
41. Which f the fllwing are crrect statements abut the differences between a machine learning algrithm and a traditinal rule-
based methd?
A. Atraditinal rule-based m ethd uses explicit prgram ening t reslpve
B. Rules in a traditinal rule-based methd can be manually and explicitly clarified,
C. Mapping relatinships f mdels in machine learning must be implicit,
D. Mapping relatinships f mdels in machine learning are autrnatically leamed.
42. Which f the fllwing statements abut the k value in the KNN algrithm are crrect?
A. The K value is a hyperparameter.
B. A larger K value indicates higher vulnerability f a mdel t verfitting.
138
C. A larger K value indicates a smther segmentatin surface f class.

D. Thek value can be set t 0.
43. In the randm frest algrithm, which f the fllwing plicies are used by a final integratin mdel t determine a mdel result?
A. Vting
B. Accumulatin
C. Averaging
D. Multiplicatin
44. Which f the fllwing can be the basis fr nde divisin in a decisin tree?
A. Gini cefficient
B. ID3
C. Infrmatin entrpy
D. CART
45. Which f the fllwing mdels can be used fr classificatin?

A. Linear regressin
B. SVM
C. Lgistic regressin
D. Decisin tree
46. If an enterprise has a large amunt f sales data withut labels. which f the fllwing mdels are suitatte fr the enterprise t
Klentify VIP custmers?
A. K-Mean
C. csv
D. Lgistic regressin
47. Which f the fllwing are tpics f speech prcessing research?

A. Speech recgnitin
B. Speech prcessing
C. Vice wakeup
D. Viceprint recgnitin
48. Which af the fllwing statements abut TensrFiw 2.x are crrect?
A. Eawy t use: The graph and sessin mecharéwns are remved.
B. The Cre functin f TensrFlw 2x is the dynamic graph mechanim called sager executin.
C. Multiple platfrms and tanguagrs are supprted, and cmpatibility b Pp Can be imprved via in n exchs fi and
D. TermurFlw 2.x desr't prvide 4 mdule enabling cmpatibility with TensrFlw 1x any mve
49. Which f the fllwing statements abut cmmn mdules f tensrflw 2.x is true?
A. tfadd :adding elements ne by ne
B. tf.data: implements peratins n datasets
C. tfdistributins: implements varius statistical distributins.
D. tfcncat:cncatenating tensrs.
50. Which f the fllwing make TensrFlw the mainstream framewrk f deep learning?
A. GPU parallel acceleratin
B. Autmatic derivatin
C. Backend develpment
D. Deep learning APIs
51. A generative adversarial netwrk is tke garne system, A generatr generates frged samples, and 4 discriminatr
determines wheter the samapre tlruee r sfal se What is the ideal result?
A. The discriminatr cannt the samples ge by the gi are trur feals e
B. The dicrimicann eaffticienrtl y the samples g by the ge are tuF efals e.
C. The generatr prduces a large number f similar samples.
D. The distibutin f generated samples fits the distributin f reat samples.
139
2. As an imprtanpatrt f a neural netwrk, activatin functarei clasnsifsie d int varius types. Which f the fllwing functins are
activatin functins f the neural
netwrk?
A. Sigmid functin
B. tach functin
C. Rell functin
D. SttPius hanctin
53. Which f the fllwing methds can be used t reslve data imbalance prblems in deep learning tasks?
A. Randm undersampling
B. Randm versampling
D. Batch deletin
54. Regulanzatin 6 an imprtant and etfective technlgy fr reducing generalizatin errrs in traditinal machine leaning. Which f
the fllwing are requiarizatin
technlgies?
A. Ut tequtarizatin
B. 2 regutatzatin
C. Srput
D. Mmentum ptimizer
55. Which f the fllwing are Al chip categries supprted frm the perspective f technical architecture?
A. CPU
B. GPU
C. ASIC
D. FPGA
56. What are the advantages f MindSpre?

A. Pipeline parallelizatin
B. In-depth graph ptimizatin
C. n-device executin
D. Device-clud cllabratin
57, Which f the fllwing statements abut MindSpre are crrect?

A. Lwer barrier fr Al develpment
B. Higher cmputing pwer
C. Efficient develpment
D. utstanding perfrmance
58, Which f the fllwing services belng t the Huawei El service family?
A. El big data services
B. Basic El services
C. Cnversatinal bt service CBS)
D. Natural language prcessing
59. Which f the fllwing capabilities are prvided by gene knwledge graphs?
A. Entity query
B. Assisted diagnsis
C. Gene detectin reprt generatin
D. Disease predictin and diagnsis
60. Which f the fllwing smartphne mdels are supprted by HUAWEI HiAl?
A. Mate 20
B. P30
C. iPhne10
D. HNR View20
61. A loss function of logistic regression is a cross entropy loss function.

140
A. True
B. False
62. Regular expressions can be added to logistic regression to avoid overfitting.

A. True
B. False
63. Grid search is a parameter tuning method.

A. True
B. False
64. Federated learning uses different data sources to train models, further breaking data bottlenecks while ensuring data
A. True
B. False
65. Computer vision is a science of how to make computers

A. True
B. False
66. The current PyTorch version supports Python 2 and Python 3.

A. True
B. False
67. The graph and session mechanism is deleted from TensorFlow 2.0.
A. True
B. False
A. True
B. False
69. The softmax function is often used as the output layer of a multiclass classification task.
A. True
B. False
70. All convolutional kernels f the same convolutional layer on a convolutional neural network share a weight.
A. True
B. False
71. GPU is good at computing intensive and parallel programs.

A. True
B. False
72. The typical application development process of the application development service des not include data lading.
A. True
B. False
73. Mind Spare’s vision and value are t lower the threshold for Al development in the industry, release the computing power
f Ascend chips, and facilitate inclusive Al
A. True
B. False
74. Knowledge graph can be applied to social networking, gene sequencing, and risk control.
A. True
B. False
A. True
141
B. False
76. What is the most important difference among batch gradient descent, mini-batch gradient descent, and stochastic
gradient descent?
A. Leaning rate
C. Batch size
D. Gradient size
77. For multiclass problems, softmax assign probability values t each class. The sum of these probability values is ().
A. 0.1
B. 0.5
C. 1
D. 100
78. Which of the following is an incorrect statement about the GBDT algorithm compared with the random forest algorithm?
A. GBDT algorithm is more prone t underfitting than the random forest algorithm.
B. The GBDT algorithm is more prone to underfitting than the random forest algorithm
C. The random forest algorithm supports parallel computing while the GBDT algorithm doesn’t
D. Both the GBDT algorithm and the random forest algorithm are based on CART trees.
79. When was the concept f Al first proposed?

A. 1950
B. 1956
C. 1946
D. 1959
80. What is the function of keras.preprcessing f TensorFlw 2.0?

B. Keras mdel generation tool
C. Keras mdel deployment tool
81. Which of the following doesn’t support dimension conversing in TensorFlow 2.0?
A. gather
B. reshape
C. transpose
D. squeeze
82. Which of the following is not a feature f TensorFlow 2.0?

A. Multi-lingual
B. Multi-platform
C. Distributed
83. Which of the following methods is used for network compilation when the Keras interface f TensorFlow 2.0 is used to
build 4 neural networks?
A. fit
B compile
C write
D. join
84. Which f the following is not included in built-in metrics f keras.metrics in TensorFolw 2.0?
A. mean
B. Accuracy
C. Recall
D. Sum
85. Which f the following optimizers can automatically adjust the learning rate?
142
A. SGD algorithm
B. Mini-batch gradient descent algorithm
C. Momentum optimizer
D. AdaGrad optimizer
86. In which f the following cases, you can do early stopping?

A. When the loss values of the training set and validation set increase at the same time
B. When the loss value of the training set decreases and the loss value of the validation set increases
C. When the loss values of the training set and validation set decrease at the same time
D. When the loss value of the training set increases and the loss value of the validation set decreases
87. Which of the following statement about gradient descent is correct?

A. The learning rate should be consistent when gradient descent is used.
B. The convergence process of stochastic gradient descent is more stable than that f global gradient
descent.
C. Stochastic gradient descent is sensitive to digital noise.
D. Adam is not a gradient descent algorithm.
88. Which of the following statements about L1 and L2 regularization is correct?

A. Compared with adding L1 norm, adding L2 norm makes the model parameters sparser.
B. the L1 and L2 regular terms cannot work on loss functions.
C. The gradient descent algorithm cannot be used to iterate parameter values after the L2 regular term is added.
D. the L1 regular term helps to enhance the generalization capability of models.
89. Which layer of the network uses the Softmax function in VGG16?
B. Poling layer
C. output layer
D. all of the above
90. Which of the following statements about a Generative Adversarial NetwrksGAN) is incorrect?
A. The generative adversarial network includes a generative model and a discriminative model.
B. The generative adversarial network cannot be used to generate new images.
C. Some discriminators require labeled samples.
D. The input of the generative model can be random numbers.
91. An objective of neural network training is to reduce 2 loss function, Which f the following methods is Commonly used to
A. Dropout
B. Gradient descent
C. Normalizer
92. Which of the following statements about hidden layers of neural networks is correct?
A. The number of hidden layers has no effect on the expressing ability of the neural network.
B. The more hidden layers, the easier the neural network will converge.
C. The more hidden layers, the easier the vanishing gradient will be caused.
D. The more hidden layers, the better the performance f the neural network.
93. Which f the following statement about the gradient descent algorithm is correct?
A. The global gradient descent algorithm must can find the minimum value of a loss function.
B. Gradient descent is an optimization algorithm t find local optimal value.
C. The stochastic gradient descent algorithm must can find the minimum value of a loss function.
D. The batch gradient descent algorithm must can resolve problems related to the local minimum value.
94. Which of the following statements about backpropagating is incorrect?

A. Backpropagation means that errors are propagated in backward direction through a network.
B. Backpropagation is used to update the weights in neural network.
C. Backpropagation can only be used on a feedforward neural network.
143
D. Backpropagation can collaborate with a gradient descent algorithm t update a network weight.
95. Which of the following does not belong t a Huawei full-stack solution?
A. Chip enablement
B. Open-source framework
C. Application
D. Edge computing
96. Which f the following attributes is the same for the Ascend 310 and Ascend 910?
A. Architecture
B. Maximum power
C Integer precision
D. 7nm process
97. Which f the following is not a feature of the Mind Spore core architecture?
A. Auto differentiating
B. Auto parallelization
C. Auto ending
D. Auto tuning
98. Tensor is a data storage component in MindSpore.

A. True
B. False
99. An &-commerce company always has issues such as poor network resource utilization and difficult ORM,, Which f the
following EIi services can be used by the company?
A. Campus Intelligent Twins
B. Industrial intelligent Twins
C. Traffic Intelligent Twins
D. Network Intelligent Twins
100. Which f the following modules provide face detecting n the HUAWEI HiAl platfrm?
A. HiAl Framework
B. HiAl Service
C. HiAl Foundation
D. HiAl Engine
101. In which f the following cases should machine learning is used t resolve actual problems?
A. Rules are complex or cannot be described, such as facial recognition and speech recognition.
B. The rules of the task will be changed with time, such as part-f-speech tagging, new words r meanings will be generated
at any time.
C. Data distribution changes over time, requiting constant readaptation f programs, such as forecasting the trend f
commodity sales.
D. Problems have low complexity and a small scale.
102. Which of the following are common types of dirty data?

A. Duplicate values
B. Missing values
C. Values in incorrect formats
D. Logically incorrect values
103. In the random forest algorithm, which f the following policies are used by a final integrating model determine a model
result?
A. Voting
B. Accumulation
C. Averaging
D. Multiplication
104. Which of the flowing are machine learning algorithms?

144
A. Supervised learning
B. Unsupervised learning
C. Semi-supervised learning
D. Reinforcement learning
105. Which of the flowing options are the advantages feature selecting?
A. Simplifying models for easier interpretation
B. Reducing the training time
C. Avoiding curse f dimensionality
D. Improving model generalization and avoiding overfitting
106. Which f the following pertains are mandatory when machine learning is used to build a model?
A. Data acquisition
B. Feature selecting
C. Model building
D. Hyper parameter tuning
107. Which of the following are Huawei's full-stack Al solutions?

A. ModelArts
B. MindSpore
C. CANN
D. Ascend
108. Which of the following TensorFolw versions have been released by the end of 2019?
A. 01
B.1
C.2
D.3
109. Which of the following make TensorFolw the mainstream framework f deep learning?
A. GPU parallel acceleration
B. Automatic derivation
C. Backend development
D. Deep learning APIs
110. Which of the flowing statements about TensorFolw 2.x are correct?
A. Easy to use the graph and session mechanizes are removed.
B. The core function f TensorFolw 2x is the dynamic graph mechanism called eager execution
C. Multiple platters and languages are supported, and compatibility between components can be improved via
standardization on exchange formats and algorithms of APIs.
D. TensorFolw 2.x doesn’t provide 4 module enabling compatibility with TensorFolw 1c any more.
111. which f the following structures belong t LSTM?

A. Input gate
B. Forget gate
C output gate
D. Memory gate
112. Currently, generative adversarial networks are widely used. Which of the following scenarios can use such networks?
A. image generation
C. Data augmentation
D. information retrieval
113. Deep learning is a popular Al technology. Various platforms may occur in deep learning tasks. Which of the following
problems may occur deep learning tasks?
A. Data imbalance problem
B. Overfitting problem
C. Vanishing gradient problem
145
D. Exploding gradient problem
114. Which f the following can a perceptron be displayed as in space?

A. Pint
B. Line
C. Plane
D. Hyperplane
115. Which of the following are Al chip categories supported from the perspective of technical architecture?
A CPU
B. GPU
C. ASIC
D. FPGA
116. Which of the following key technologies are used by MindSpore t support device-edge-cloud n-demand collaborating?
A. Unified model IR
B. Graph optimization technology based on software and hardware collaboration
C. Federal Meta learning based n device-cloud collaboration
D. Centralized architecture
117. Which two of the following are included in MindSpore?

A. TBE (operator development)
B. Mind Expressing
C. Graph Engine
D. Third-party framework TF/Cafes
118. Which of the following services belong t the Huawei El service family?
A. El big data services
B. Basic El services
C. Conversational bot service CBS)
119. Which of the following are features of the intrusion detecting service provided by HUAWEI CLUD El?
A. High flexibility
B. Ease-of-use
C. Low misreporting rate
D. High labor cost
120. Which of the following smartphone models are supported by HUAWEI HiAl?
A. Mate 20
B. P30
C. iPhne10
D. HNR View20
121. The training error decreases as the model complexity increases.

A. True
B. False
Answer: A
122. About the following code: SVR I kernel=" r b f’, C=Te3, gamma=0.01), it means that we use linear kernel function.
A. True
B. False
Answer: A
123. If the L1 regular expression is added to the loss function of linear regression, the regression is called Lasso
regression,
A. True
B. False
Answer: A
146
124. Federated learning uses different data sources to train models, further breaking data bottlenecks while ensuring data
A. True
B. False
Answer: B
125. By “all Ab scenarios’, Huawei means different deployment scenarios for AI including public Clouds, private clouds,
edge computing in all forms industrial loT devices, anet Consumer devices.
A. True
B. False
Answer: A
126. The current PY Torch version supports Python 2 and Python 3.

A. True
B. False
Answer: B
127. Tensor Flow 2.0 supports GPU acceleration, That is, parallel operation acceleration is supported for matrix addition,
subtraction, multiplication, and division.
A. True
B. False
Answer: B
128. All convolutional kernels of the same convolutional layer on a convolutional neural network share a weight.
A. True
B. False
Answer: A
129. The soft max function is often used as the output layer of a multiclass classification task.
A. True
B. False
Answer: A
130. The training principle used by a recurrent neural network is called back propagation through time (BPTT).
A. True
B. False
Answer: A
131. Al chips, also known as Al accelerators, ate function modules used to process massive computing tasks in Al
applications.
A. True
B. False
Answer: A
132. Mind Spore can use only Huawei-developed Ascend processors for training and inference.
A. True
B. False
Answer: A
133. Mind Spore supports Al model training and inference based on Huawei-developed Ascend processors, it can run in
various distributed environments such as single-node and multi-card and multi-node and multi-card.
A. True
B. False
Answer: A
A. True
B. False
Answer: A
147
135. HUAWEI HiAl Engine can easily integrate multiple artificial intelligence (Al) capabilities with apps.
A. True
B. False
Answer: A
136. What is the relationship between base learners in Bagging integrated learning?
A. Mutually independent
B. The latter model must be built on the former one.
C. Correlated relationship
D. Additive relationship
137. Which of the following is a good state for a machine learning evaluation model?
A. High bias & high variance
B. High bias & low variance
C. Low bias & high variance
D. Low bias & low variance
Answer: both B and C
138. Which of the following is the correct statement about the overall process of machine learning?
A. Data collection » Data cleansing » Feature extraction and selection » Model training » Model evaluation and test »
Model deployment and integration
B. Data collection » Data cleansing » Model training » Feature extraction and selection » Model evaluation and test »
C. Data collection > Feature extraction and selection > Data cleansing > Model training > Model evaluation and test >
D. Data collection » Data cleansing > Feature extraction and selection > Model training > Model deployment and
integration » Model evaluation and test
Answer: C
139. Which of the following schools does neural network research belong to?
A. Symbolism
B. Connectionism
C. Behaviorism
Answer: B
140. Which of the following methods can be used for tensor combination in TensorFlow 2.0?
A. concat
B. unstack
C. split
D. join
Answer: A
141. Which of the following module is applied to save the Keras model weights at some frequency during model training?
A. keras.callbacks
B. tfkeras.models
C. keras.utils
D. tikeras.initializer
Answer: A
142. Which of the following companies first develops Tensor Flow?

A. Oracle
B. Facebook
C. Google
D. NVIDIA
Answer: C
143. Which of the following data types is used by Tensor Flow?

148
A. Scalar
B. Vector
C. Matrix
D. Tensor
Answer: D
144. Which of the following does not support tensor creation in Tensor Flow 2.0?
A. fill
B. create
C. constant
D. zeros
Answer: D
145. In 1958, Rosenblatt invented the perception algorithm, which was used for about 10 years. In 1969, Minsky
questioned the capability of the perceptron to resolve non
finear problems and proposed the famous prothom that slows down the development of perceptron. The protein is an ( ).
A. AND problem
B. OR problem
C. XOR problem
D. XAND problem
Answer: C
146. Which of the following statements about back propagation is incorrect?

A. Back propagation means that errors are propagated in backward direction through a network.
B. Back propagation is used to update the weights in neural network .
C. Back propagation can only be used on a feed forward neural network.
D. Back propagation can collaborate with a gradient descent algorithm to update a network weight.
Answer: C
147. Dataset expansion can effectively affect over fitting occurrence. Which of the following statements about dataset
expansion is correct?
A. A larger dataset indicates a lower over fitting probability.
B. A larger dataset indicates a higher over fitting probability.
C. A smaller dataset indicates a lower over fitting probability.
D. An enlarged of shrunk dataset indicates a lower over fitting probability.
Answer: D
148. A loss function reflects an error between 4 target output are not an actual output of a neural network. Which of the
following a common loss function in deep learning?
A. Logarithmic loss function
B. Exponential loss function
C. Hinge loss function
D. Mean square loss function
Answer: D
149. Which of the following statements about optimizers is correct?

A. The learning rate of the momentum optimizer does not need to be set manually.
B. The learning rate of the Ada Grad optimizer is not automatically updated.
D. There is no relationship between the Adam optimizer and Ada Grad optimizer.
Answer: C
150. An objective of neural network training is to reduce 2 loss function. Which of the following methods is commonly used
to minimize a loss function?
A. Dropout
B. Gradient descent
C. Nonnalizer
Answer: B
149
A. Hyper parameter
B. Feature value
D. Parameters
Answer: D
152. Which layer of the network uses the Softmax function in VGG16?
B. Pooling layer
C. Output layer
D. All of the above
Answer: Both A and B
153. What does mini-batch mean in mini-batch gradient descent?

A. During each parameter iteration, some data is extracted from datasets for training.
B. During each parameter iteration, some data is extracted from features for training.
C. During each parameter iteration, some data is extracted from errors for training.
D. During each parameter iteration, some data is extracted from hyperparameters for training.
Answer: C
154. Which of the following statements about gradient descent is correct?

A. The learning rate should be consistent when gradient descent is used.
B. The convergence process of stochastic gradient descent is more stable than that of global gradient descent.
C. Stochastic gradient descent is sensitive to digital noise.
D. Adam is not a gradient descent algorithm.
Answer: D
155. How many layers does the software stack of an Ascend Al chip consist of?
A. 3
B. 4
C. 2
D. 5
Answer: B
156. Which of the following attributes is the same for the Ascend 310 and Ascend 910?
A. Architecture
B. Maximum power
D. 7nm process
Answer: Both A and B
157. Which of the following is the operation type of nn?

A. Mathematical calculation
B. Network
C. Control
D. Others
Answer: B
158. Which of the following statements about the architecture of MindSpore is not true?
A. Auto Differ
B. Auto Parallelism
C. Auto tuning
D. Auto training
Answer: B
159. Which of the following products can be used by a company to enable facial recognition access control?
150
A. Model Arts
B. HiLens
C. OBS
D. ECS
Answer: A
160. Which of the following tools can be integrated with the HUAWEI HiAl platform?
A. MyEclipse
B. Android Studio
C. Spider
D. Jupyter Notebook
Answer: B

A. Duplicate values
B. Missing values
Answer: A,B and C
162. If an enterprise has a large amount of sales data without labels, which of the following models are suitable for the
enterprise to identify VIP customers?
A. K-Means
B. Hierarchical chistering
C.SVM
D. Logistic regression
Answer: D
163. From sklearn.linear_ model, we can import () modules?

A. RidgeCv
B. LassocV
C. LinearRegression
D. SVM
Answer: C
164. Which of the following models can be used for classification?

A. Linear regression
B. SVM
C. Logistic regression
D. Decision tree
Answer: A and C
165. Which of the following can be the basis for node division in a decision tree?
A. Gini coefficient
B. ID3
C. Information entropy
D. CART
Answer: C
166. Which of the following are common clustering algorithms?

A. K-means
C. Density-based clustering
D. Spectral clustering
Answer: B and C
167. Which of the following are Al application fields?
A. Smart home
B. Smart healthcare
C. Smart city
151
D. Intelligent education
Answer: A
168. Which of the following Tensor Flow versions have been released by the end of 2019?
A. 0.1
B. 1
C. 2
D. 3
Answer: C
169. Which of the following steps belong to Tensor Flow development process?
A. Data preparation
B. Network construction
C. Model training and verification
D. Model restoration and invoking
Answer: A and C
170. Which of the following methods can be used to check whether a data type is tensor in Tensor Flow 2.0?
A. is tensor
B. isinstance
C. device
D. dtype
Answer: D
171. In a convolutional neural network, different layers have different functions. Which of the following layers can reduce
dimensions?
B. Pooling layer
C. Fully-connected layer
D. Input layer
Answer: B
172. A pooling layer in a convolutional neural network can reduce the size of feature maps. Which of the following are
Common pooling layers?
A. Average-pooling layer
B. Max-pooling layer
C. Min-pooling layer
D. Product pooling layer
Answer: A and B
173. Which of the following steps in deep learning are automatically performed by models?
A. Task analysis and locating
B. Model training
C. Feature extraction
D. Feature selection
Answer: B
174. Which of the following methods can be used to resolve data imbalance problems in deep learning tasks?
A. Random under sampling
D. Batch deletion
Answer: A and B
175. Al Chips can be divided into ( ) by technical architecture?

A. CPU
B. GPU
C. ASIC
D. FPGA
152
Answer: B
176. MSIR is a simple, efficient, and flexible graph based functional IR, Which of the following functional semantic types
can be represented by MSIR?
A. Free variable
C. Recursion
D. Interrupt
Answer: A and D
177. Which of the following are Al training and inference frameworks?

A. Tensor Flow
B. PyTorch
C. MindSpore
D. Pandas
Answer: A and B
178. HUAWEI CLOUD Et is an enabler of enterprise intelligence. It provides 2 Goud service platform for enterprises based
on Al and big Gata technologies. What are the
features of this platform?
A. Developed
B. trusted
C. Intelligent
D. Secure
Answer: B
179. HUAWEI CLOUD Model Arts is a one-stop Al development platform, Which of the following functions can be provided
by Model Arts?
A. Data governance
B. Automatic earning
C. Visualized workflows
D. Al market
Answer: A
180. What are the advantages of the HiAl mobile computing platform?
A. Comprehensive documents
B. Source code enabling quick start
C. Different types of APIs
D. Various tool chains
Answer: A,C and D
181. The naive Bayes algorithm does not require sample features to be independent and identically distributed.
A. True
B. False
Answer: False
182. The K-fold cross-validation is to divide a test dataset into K sub-datasets.

A. True
B. False
Answer: True
183. Neural networks are not a type of machine learning.

A. True
B. False
Answer: False
184. Algorithmic biases are mainly caused by data biases.

A. True
B. False
153
Answer: True
185. Jobs that are highly repetitive and require less on social skills are most likely to be replaced by AI.
A. True
B. False
Answer: True
186. TensorFlow is an end-to-end open-source platform for machine learning and deep learning.
A. True
B. false
Answer: True
187. In Keras, model.summary () is used to print network information.

A. True
B. false
Answer: True
A. True
B. False
Answer: True
A. True
B. False
Answer: True
190. The training principle used by a recurrent neural network is called backpropagation through time (BPTT).
A. True
B. False
Answer: True
191. The CPU improves AI performance by adding instructions and cores.

A. True
B. False
Answer: True

A. True
B. False
Answer: True
193. MindSpore supports AI model training and inference based on Huawei developed Ascend processors. It can run in
various distributed environments such as single- node multi-card and multi-node multi-card.
A True
B. False
Answer: True
194. ModelArts is a one-stop development platform tor AI developers. With data preprocessing, semi-automatic data
labeling, large-scale distributed training, automatic modeling, and on-demand model deployment on devices, edges, and
clouds. ModelArts helps AI developers build models quickly and manage the AI development lifecycle.
A. True
B. False
Answer: True
A. True
B. False
Answer: True
154
196. Which of the following algorithm in gradient descent algorithms has the most disordered track on the curved surface of
a loss function
A. BGD
B. SGD
C. MBGD
D. MGD
Answer: B
197. Which of the following is a good state for a machine learning evaluation model?
A. High bias & high variance
B. High bias & low variance
C. Low bias & high variance
D. Low bias & low variance
Answer: D
198. In machine learning, () needs to be input to a model to train the model and make predictions.
A. Artificial programs
B. Historical data
C. Neural networks
D. Training algorithms
Answer: D
199. The three phases of AI include computing intelligence, ( ), and cognitive intelligence.
A. Behavior intelligence
C. Narrow AI
D. Strong AI
Answer: B
200. Which of the following functions is not provided by PyTorch?

A. GPU acceleration
B. Dynamic graphs
C. Embedded Keras
D. Automatic derivation
Answer: C
201. Which of the following does not support dimension conversion in TensorFlow 2.0?
A. gather
B. reshape
C. transpose
D. squeeze
Answer: D
202. Which of the following statements about PyTorch ts not true?

A. Pytorch provides tensors that support CPUs and GPUs, greatly accelerating computing,
B. Pytorch can generate dynamic graphs during execution. Developers can stop an interpreter in a debugger and view
output of a specific node.
C. PyTorch programs can't dynamically build and adjust computational graphs during execution,
D. In addition to Facebook, institutes such as Twitter, GMU, and Salesforce also use PyTorch.
Answer: D
203. Which company first launched PyTorch?

A. Google
B. Huawei
C. Baidu
D. Facebook
Answer: D
204. Which of the following is not a feature of TensorFlow 2.0?

155
A. Multi-lingual
B. Multi-platform
C. Distributed
Answer: A
205. LSTM is developed based on recurrent neural networks. Which of the following problems is resolved by LSTM for
recurrent neural networks?
A. Overfitting problem
B Vanishing gradient problem
C Underfitting problem
Answer: B
206. Which of the following statements about a standard RNN model is correct?
A. Information attenuates after long-term memory transmission.
B. The many-to-many model structure does not exist.
C., The one-to-one model structure does not exist.O
D. The time orientation is not considered during back-propagation
Answer: A
207. Which of the following statements about hidden layers of neural networks is correct
A. The number of hidden layers has no effect on the expression ability of the neural network.
B. The more hidden layers, the easier the neural network will converge.
C. The more hidden layers, the easier the vanishing gradient will be caused.
D. The more hidden layers, the better the performance of the neural network.
Answer: C
208. Which of the following statements about L1 and L2 regularization is correct?

A. Compared with adding L1 norm, adding L2 norm makes the model parameters more sparse.
B. The L1 and L2 regular terms cannot work on loss functions.
C. The gradient descent algorithm cannot be used to iterate parameter values after the L2 regular term
is added
D. The L1 regular term helps to enhance the generalization capability of models.
209. Which of the following statements about activation functions is correct?

A. The Sigmoid is a linear activation function.
B. The ReLU is derivable everywhere.
C. The softmax can be used only at the output layer.
D. The tanh is a nonlinear activation function.
210. Deep learning neural network training requires a large amount of matrix computation. Generally, hardware is required
for a computer to be capable of parallel computing. Which of the following hardware devices can provide the parallel
computing capability?
A. CPU
B. DMM
C. GPU
D. Mainboard
A Hyperparameter
B. Feature value
D. Parameter
212. In 1958, Rosenblatt invented the perceptron algorithm, which was used for about 10 years. In 1969. Minsky
questioned the capability of the perception to resolve non-linear problems and proposed the famous problem that slowed
156
down the development of perceptron. The problem is an ( ).

A. AND problem
B. OR problem
C. XOR protler
D. XAND problem
213. The Softmax activation function can map any k-dimensional real number vector to another k-dimensional real number
vector, Which layer of the neural network does this function usually apply to?
A, Input layer
B. Hidden layer
D. Output Layer
214. Traditional machine learning and deep learning are core AI technologies, and their engineering processes are slightly
different. Which of the following steps does not need to be performed in deep learning?
A Data cleaning
C. Model building
D. Model evaluation
215. Which of the following statements is Ascend AI Computing Engine?

A. AI CPU
B. DVPP
C. Da Vinci Architecture
D. Buffers
216. How many layers does the software stack of an Ascend AI chip consist of?
A. 3
B. 4
C. 2
D. 5
217. The core of an AI framework, that is, one of decisive factors of the programming paradigm is the automatic
differentiation technology used by the AI framework. Which of the following method is used by MindSpore?
A. Graph method
B. Operator overloading
C. Source code conversion
218. Which of the following is the operation type of nn?
A. Mathematical calculation
B. Network
C. Control
D. Others
219. Which of the following of HUAWEI CLOUD EI can integrate AI technologies into application scenarios of various
industries and fully utilizes advantages of AI technologies to improve efficiency and experience?
A. EI intelligent Twins
B. EI big data services
C. Object storage service
D. Cloud database
220. Which of the following tools can be integrated with the HUAWEI HiAI platform?
A. MyEclipse
B. Android Studio
C. Spider
D. Jupyter Notebook
221. In which of the following cases should machine learning be used to resolve actual problems?
157
A. Rules are complex of cannot be described, such as facial recognition and speech recognition.
B. The rules of the task will be changed with time, such as part-of-speech tagging, new words of
meanings will be generated at any time.
C. Data distribution changes over time, requiring constant readaptation of programs, such as forecasting
the trend of commodity sales.
D. Problems have low complexity and a small scale.
222. Which of the following options are the advantages of feature selection?
C. Avoiding curse of dimensionality
223. Which of the following statements about the k value in the KNN algorithm are correct?
A. The K value is a hyperparameter.
B. A larger K value indicates higher vulnerability of a model to overfitting.
C. A larger K value indicates a smoother segmentation surface of class.
D. The k value can be set to 0.
224. The F1 value for evaluating classification models is the harmonic mean of indicators including ( ).
A. Accuracy
B. Precision
C. Recall
D. Validity
225. Which of the following operations are mandatory when machine learning is used to build a model?
A. Data acquisition
C. Model building
D. Hyperparameter tuning
226. Which of the following may be the segmentation surface of the support vector machine (SVM) machine learning
algorithm?
A. Straight line
B. Plane
C. Hyperplane
D. Surlace
227. Which of the following are AI application fields?

A. Smart home
B. Smart healthcare
C. Smart city
228. Which of the following TensorFlow versions have been released by the end of 2019?
A. 0.1
B.1
C. 2
D.3
229. If a model has a high accuracy on the training set but low accuracy on the test set, what methods
can be used to improve the model?
A. Increasing the data volume of the training set
B. Adding a regularization term after the1oss function
C. Dropout method
D. Early stopping
230. Which of the following libraries are used to process images and visualization toolkits?
A. Matplotlib
158
B. scikit-image
C. Opencv
D. scikit-learn
231. Which of the following are commonly used optimizers in deep learning model training?
A. SGD
B. Momentum
C. Adam
D. AdaGrad
232. As an important part of a neural network, activation functions are classified into various types. Which of the following
functions are activation functions of the neural network?
A. Sigmoid function
B. Tanh function
C. ReLU feretion
D. Softplus function
233. In a convolutional neural network, different layers have different functions, Which of the following layers can reduce
dimensions?
B. Pooling tayer
C Fully-connected layer
D. input layer
A. Random undersampling
D. Batch deletion
235. Which of the following Atlas products are used in training scenarios?
A. Atlas 300 (model 9000)
B. Atlas 500
C. Atlas 800 (model 9000)
D. Atlas 900
236. MSIR is a simple, efficient, and flexible graph-based functional IR. Which of the following functional semantic types
can be represented by MSIR?
A. Tree variable
C. Recursion
D. Interrupt
237. Which of the following are common tensor operations in MindSpore?

A. asnumpy0
B. size()
C. switch()
D. tensor_add(other:Tensor)
238. Which of the following are correct statements about the product advantages of the ModelArts AI evelopment platform?
A. One-stop platform
B. Ease of use
C. High performance
D. Flexibility
239. Which of the following capabilities are provided by gene knowledge graphs?
A. Entity query
B. Assisted diagnosis
C. Gene detection report generation
159
D. Disease prediction and diagnosis
240. Which of the following smartphone models are supported by HUAWEI HiAl?
A. Mate 20
B. P30
C. iPhone10
D. HONOR View20
241. Grid search is a parameter tuning method.

A. True
B. False
Answer: True
242. Aloss function of logistic regression is a cross entropy loss function.

A. True
B. False
Answer: True
243. If the LT regular expression is added to the loss function of linear regression, the regression is called Lasso
regression.
A. Tue
B. false
Answer: True
244. Federated learning uses different data sources to train modeis, furthet breaking data bottlenecks while ensuring data
A. True
B. False
Answer: False
245. For neural networks that are based on connectionism, each node can express a specific meaning.
A. True
B. False
Answer: True
246. The graph and session mechanism is deleted from TensorFlow 2.0.
A. True
B. False
Answer: True
247. The tficontrib method can still be used in TensorFlow 2.0.

A. True
B. False
Answer: True
A. True
B. False
Answer: False
A. True
B. False
Answer: True
A. True
B. False
Answer: True
160
251. Al chips, also known as Al accelerators, are function modules used to process massive computing tasks in Al
applications.
A. True
B. False
Answer: False

A. True
B. False
Answer: True
253. MinctSpore supports Al modet training and wiference based on Huawe-developed Ascend processors. Mt can fun in
various distributed ervitonments such as single-node
multh-card and multi-ode multi-card.
A. Two
B. Five
A. True
B. False
255. The HIAI mobile computing platform provides various upper-layer function service APIs to ensure efficient running on
mobile devices,
A. True
B. False
256. Which of the following is the label predicted by the regression algorithm?
A. Discrete
B. Continuous
C. Independent
D. Dependent
257. In python, math.pow is for ()?

A. Square
B. Take any power
C. Summation
D. Quadrature
258. Which of the following statements about the SVM algorithm is incorrect?
A. The SVM algorithm cannot be used to process non-linear datasets.
B. The SVM algorithm can use multiple kernel functions.
C. Asupport vector is a vector composed of points nearest to the hyperplane.
D. SVMs can be used to resolve classification problems.
259. When was the concept of Al first proposed?

A. 1950
B. 1956
C. 1946
D. 1959
C. Preventing overfitting
Answer: C

161
A. Google
B. Huawei
C. Baidu
D. Facebook
Answer: D
262. Whach of the following methods is used for network compilation when the Keras interface uf TensorFlow 2.0 is used
to build @ neural network?
A. Fit
B. & compile
C. write
D. Pm
Answer: B
263. Which of the following data types is used by TensorFlow?

A. Scalar
B. Vector
C. Matrix
D. Tensor
Answer: ABCD
264. Which of the following methods can be used to view tensor dimensions in TensorFlow 2.0?
A. dimens
B. dtype
C. device
D. ndim
Answer: D
265. in deep learning, the vanishing gradient problem is fikety to occur if there are 4 large number of neural network layers.
Strictty speaking. in which of the following phases
does the vanishing grad&ent problem occur?
A. Result calculation in formant propagation
B. Parameter update in backpropagation
C. Parameter update in forwant propagation
D. 0. Rest calculation in backpropagation
C. The one-to-one model structure does not exist.
D. The time orientation is not considered during backpropagation.
267. How should samples in real datasets be used in Generative Adversarial Networks(GAN)?
A. Used as the input value of the generative model
B. Used as the output value of the generative model
C. Used as the input value of the discriminative model
D. Used as the output value of the discriminative model
268. Perceptron is the simplest neural network among all deep learning neural networks, Which af the followeng
statements about the perceptran’s structure is correct?
A. The perceptron has no hidden layer.
B. The perceptron uses the sigmoid activation function
C. The perceptron has two hidden layers.
D. The percegtron uses the Ret activation function

A. Momentum accelerates the model training process.
B. When Adagrad is used, the learning rate remains unchanged.
C. When RMSprop is used, the learning rates of all parameters are the same.
162
D. The Adam optimizer does not need to set the initial learning rate.
270. Regularization is often used in deep learning. What problem is resolved by regularization in deep learning?
B. Vanishing gradient problem
C. XOR problem
D. Overfitting problem
Answer: D

D. There is no relationship between the Adam optimizer and AdaGrad optimizer.
272. The Sottrax activation function can map any k-dimensional real number vector te another k-dimensional real number
vector, Which layer of the neural network does
this function usually apply to?
A. Inpat layer
B. 6 Hidden layer
D. Output layer
273. Which of the following statements about the cross-entropy loss function is correct?
A. The cross-entropy loss function is often used in classification problems
B. The cross-entropy loss function is often used in regression problems
C. The cross-entropy loss function can be used for both classification and regression problems
D. The cross-entropy loss function can only be used for binary classification problems
274. L1 and L2 requterization are common methods used by traditional machine learning to reduce generalization errors,
which of the following statements about LI and L2
regularization is correct?
A. Wregulaization can be used for feature selection,
B. Wregularization can be wed for feature selection
C. Both Lt and 12 regdtarization can be sed tor feature selection
D. Lt and (2 regularization cannot be used for feature selection.
275. How many layers does the software stack of an Ascend Al chip consist of?
A. 3
B. 4
C. 2
D. 5
Answer: B
276. Which of the following does not belong to a Huawei full-stack solution?
A. Chip enablement
B. Open-source framework
C. Application
D. Edge computing
Answer: B,D
277. Which of the following is not a common tensor operation in MindSpore?

A. asnumpy0
B. size()
C. dimd
D. ford
Answer: C,D
163
278. Which of the following processing modes is used by MindSpore to implement efficient distributed training for ultratarge
models?
A. Auto parallel
B. Manual parallel
C. Serial
Answer:
279. Traffic Intelligent Twins can implement () hours of full-time/all-domain traffic awareness.
A. 7x24
B. 5x24
C. TK12
D. 5x12
280. Which of the following HUAWEI HiAl modules is supported by Huawei chips?
A. HiAl Service
B. HiAl Foundation
C. HiAl Framework
D. HiAl Engine
281. Which of the following problems can be resolved by using machine learning algorithms?
A. Classification problem
B. Regression problem
C. Reinforcement problem
A. Data acquisition
C. Model building
283. Which of the following options are the advantages of feature selection?
A. Rules are complex or cannot be described. such as facial recognition and speech recognition.
B. The rufes of the task will be changed with time, such as part-of-speech tagging, new words or meanings will be
generated at any time. C. Data distribution changes over time.
requiring constant readaptation of programs, such as forecasting the trend of commodity sales.
D. Problems have low complexity and # small scale.
285. The F1 value for evaluating classification models is the harmonic mean of indicators including ().
A. Accuracy
B. Precision
C. Recall
D. Validity

A. Duplicate values
B. Missing values
287. What are the main technical applications of Al?

A. Computer vision
164
B. Natural language processing

C. Control system
288. It a model has a high accuracy on the training set but low accuracy on the test set, what methods can be used to
improve the model?
A. tnereasing the data volume of the training set
B. Adding a regularization term after the boss function
C Dropout method
D. Early stopping
289. Which of the following steps belong to tensorFlow development process?

A. Data preparation
B. Network construction
C. Model training and verification
D. Model restoration and invoking
290. Which of the following are network training methods in the Keras interface of TensorFlow 2.0?
A. Fit
B. Fit_generator
C. predict
D. evaluate
291. Regularization 6 an important and effective technology for reducing generalization erors in traditional machine
learning. Which of the following are requtarzation
technologies?
A. Regulation
B. 12 regularization
C. Dropout
292. In convolutional neural networks, which of the following scenarios are suitable for using pooling layers?
A. Flower recognition
B. AlphaGo
C. Medical image segmentation
D. Face recognition
293. If the vanishing or exploding gradient problem occurs on 4 deep learning neural network, which of the following are
Common solutions?
A. RetU activation function
B. Gradient shear
C. Normaticer
D. Random undersampling
294. Which of the following are commonly used loss functions in deep learning?
A. Mean squared error loss function
B. Cross-entropy error loss function C. IoU loss function
D. Mean absolute error
295. Which of the following are Al chip categories supported from the perspective of service application?
A. Training
B. GPU
C. Inference
D. Model building
296. Which of the following are common tensor operations in MindSpore?

A. asnumpyQ
B. sizeQ
C. switchd
165
D. tensor_add(other:Tensor)
297. Which two of the following are included in MindSpore?

B. Mind Expression
C. Graph Engine
298. Which of the following are correct statements about the product advantages of the ModelArts Al development
platform?
A. One-stop platform
B. Ease of use
C. High performance
D. Flexibility
299. Which of the following smartphone models are supported by HUAWEI HiAI?
A. Mate 20
B. P30
C. iPhone10
D. HONOR View20
300. HUAWEI CLOUD ET is an enabler of enterprise intelligence. It provides a cloud service pAatiorm for enterprises
based on Al and big data technologies, What are the
features of this platform?
A. Developed
B. 6BTrusted
C. Wetigent
D. Secure
301. Neural networks are not a type of machine learning.

A. True
B. False
Answer: B
302. About the following code: SVRikemet=" rbf , C=1e3, gamma=0.01), it means that we use linear kernel function.
A. True
B. False
Answer: A

A. True
B. False
Answer: B
304. The smart speaker is one of the typical application scenarios of speech processing.
A. True
B. False
Answer: A
305. At present, Al is still in the narrow Al phase.

A. True
B. False
Answer: A
306, Tensor Flow is now one of the most popular deep learning frameworks.
A. True
B. False
Answer: A
166
307. Tensor Flow is an end-to-end open-source platform for machine learning and deep learning.
A. True
B. False
Answer: A
A. True
B. False
Answer: A
A. True
B. False
Answer: A
310. An artificial neural network (ANN) is an information processing system designed to imitate the structure and functions
of human brain.
A. True
B. False
Answer: A
311. Huawei's Ascend Al chip is one of the typical NPUs.

A. True
B. False
Answer: A
312. Mind Spore’s vision and value are to lower the thineshald for Al development in the industry, release the computing
power of Ascend chips, and facilitate inclusive Al
A. True
B. False
Answer: B
313. Mind Spore can use only Huawei-developed Ascend processors for training and inference.
A. True
B. False
Answer: A
14. Knowledge graph can be applied to social networking, gene sequencing, and risk control,
True
False
Answer: A
315. HUAWEI HCAl Engine can easily integrate multiple artificial intelligence (Al) capabilities with apps.
A. True
B. False
Answer: A
316. Factory A produces 40% of an automotive company’s products, and factory B accounts for 60% The detective rate of
factory A is 1%, and that of factory Bis 2% It vehicle is defective, what is the possibility that the vehicle is produced by
factory A?
A. 0.2
B. 0.25
C. 0.3
D. 0.5
Answer: B
317. When we use this code: Grid Search CV (SVR (), param_grid=parameters, cv=3), cv=3 means()?
A. SVR iterative three times O
B. 3 classification
167
C. Do a 3 fold cross validation?

D. The data is 3d
Answer:
318. Which of the following Is the problem type for "predicting that Ihe store revenue in the next quarter is 200,000-300,000
or 400,000-500,000 based on héstorical data with fabets”?
B. Regression preform
C. Rate problem
D. Chartering problem
Answer: D
319. According to the American Society of Automotive Enginwers (SAD), which of the folowing are the lewels of
autonomous driving clasutied by the degree of dependence on the system?
A. 10-14
B. 10-15
C. 11-15
D. 11-14
Answer: B
320. Which of the following is not a feature of Tensor Flow 2.0?

A. Multi-lingual
B. Multi-platform
C. Distributed
Answer: A
321. Which of the following methods can be used to view tensor dimensions in Tensor Flow 2.0?
A. dimens
B. dtype
C. device
D. ndim
Answer: B
C. Preventing over fitting
Answer: C
323. Which of the following is not included in built-in metrics of keras.metrics in Tensor Flow 2.07?
A. mean
B. Accuracy
C. Recall
D. sum
Answer: D
324. What is the meaning of the epochs parameter in model fit(mnisttrain.image, mnist.train.labels.epochs=5)?
A. The whole training set will be divided into five parts.
B. The whole training set will be divided into six parts
C. All training sets will be trained five times.
D. All test sets will be tested five times.
Answer: B

D. There is no relationship between the Adam optimizer and AdaGrad optimizer.
168
Answer: C
326. If each layer of a neural network uses the sigmoid activation function and the number of layers is large, which of the
following problems will occur?
A. Gradient explosion problem
B. Vartshing graclient problem
C. Under fitting problem
D. Over tatting problem
Answer: B
327. Which of the following statements about the cross-entropy loss function is correct?
A. The cross-entropy loss function is often used in classification problems
B. The cross-entropy loss function is often used in regression problems
C. The cross-entropy loss function can be used for both classification and regression problems
D. The cross-entropy loss function can only be used for binary classification problems
Answer: D
328. Regularization is often used in deep learning. What problem is resolved by regularization in deep learning?
B. Vanishing gradient problem
C. XOR problem
D. Over fitting problem
Answer: D
329. Deep leaning neural network training requires a large amount of matrix computation. Generally, hardware is required
tor a Compiler to be capable of parallel computing. Which of the following hardware devices can provide the parallel
Computing capability?
A. CPU
B. DIMM
C. GPU
D. Mainboard
Answer: A
C. The one-to-one model structure does not exist.
D. The time orientation is not considered during back propagation.
Answer: D
331. Dataset expansion can effectively affect over fitting occurrence. Which of the two Bowing statements about dataset
expansion its correct?
A. A larger dataset indicates a lower over fitting probability,
B. A larger dataset indicates a higher over fitting probability,
C. A smaller dataset indicates a lower over fitting probability.
D. An enlarged of shrunk dataset indicates a lower over fitting probability
Answer: D
332. Which of the following statements about a Generative Adversarial Networks(GAN) is incorrect?
A. The generative adversarial network includes a generative model and a discriminative model.
B. The generative adversarial network cannot be used to generate new images.
C. Some discriminators require labeled samples.
D. The input of the generative model can be random numbers
Answer: C
333. Which of the following is the output range of the Sigmoid function? O
A. (0,+ꚙ)
B. (0,1)
C. E1,1]
169
D. [-1,0]
Answer: B
334. Which of the following optimizers can automatically adjust the learning rate?
A. SGD algorithm
Answer: A
335. Which of the following is not included in the logical architecture of Ascend Al processors?
A. Control CPU
B. Al computing engine
C. DVPP
D. GPU
Answer: C
336. Which of the following is the main application of the Ascend 310 chip serving as the basis of the
A. Atlas 200 DK?
B. Model training
C. Model building
D. Model inference
Answer: A
337. Mind Spore has made many optimizations to improve computing performance. Which of the following is not Mind
Spore’s optimization?
A Pipeline parallelization
B. Dedicated memory
C. On-device execution
Answer: A
338. The core of an Al framework, that is, one of decisive factors of the programming paredigm is the automatic
differentiation technology used by the Al framework. Which of the following method is used by Mind Spore?
A. Graph method
B. Operator overloading
C. Source code conversion
Answer: B
339. Which one doesn't belong to Huawei El Essential Platform?

A. Model Arts
B. Huawei HiLens
C. Graph Engine Service (GES)
D. Huawei HiAl
Answer: C
340. Which of the following HUAWEI HiAl modules is supported by Huawei chips?
A. HiAI Service
B. HiAl Foundation
C. HiAl Framework
D. HiAl Engine
Answer: B

B. SVM
D. Decision tree
Answer: B
170
342. From sklearn. linear model, we can import () modules?

A. Ridge CV
B. Lasso CV
C. Linear Regression
D. SVM
Answer: C
A. Rules ate complex of cannot be described, such as facial recognition and speech recognition.
B. The rules of the task will be changed with time, such as part-ot-speech tagging, new words or
the trend of commodity sales
D. Problems have low complexity and a small scale,
Answer: C
344, Which of the following are common types of dirty data?

A. Duplicate values
B. Missing values
Answer:A
345. Which of the following can be the basis for node division in a decision tree?
A. Gini coefficient
B. ID3
C. Information entropy
D. CART
Answer: D
346. Which of the following problems can be resolved by using machine learning algorithms?
B. Regression problem
C. Reinforcement problem
Answer: A
347. What are the application scenarios of computer vision?

A. Action analysis
B. Smart album
C. Image search
D. Voice navigation
Answer: C
348. Mf the total number of parameters at a layer of a fully-connected neural network is 330, the numbers of neurons at the
upper layer and the current layer may be (), respectively.
A. 32and 10
B. 33.and 10
C. 9 and 33
D. 10 and 33
Answer: A
349. Which of the following methods can be used to check whether a data type is tensor in TensorFlow 2.0?
A. is tensor
B. is instance
C. device
D. dtype
Answer:A
171
350. If a model has a high accuracy on the training set but tow accuracy on the test set, what methods can be used to
improve the model?
A. Increasing the data volume of the training set
B. Adding a regularization term after the loss function
C. Dropout method
D. Early stopping
Answer:B
A. Random under sampling
D. Batch deletion
Answer:A
352. Deep teaming is 4 popular Al technology; Various problems may occur in deep learning tasks. Which of the following
problems may occur in deep learning tasks?
A Data imbalance problem
B. Overfitting problem
C. Vanishing gradient problem
Answer:c
353. Which of the following steps in deep learning are automatically performed by models?
A. Task analysis and locating
B. Model training
C. Feature extraction
D. Feature selection
Answer:D
354. Which of the following statements about a convolutional neural network pooling layer are correct?
A. The pooling layer provides the function of dimension reduction.
B. Max-pooling and average-pooling are common pooling methods.
C. Featured images become smaller after pooling.
D. Pooling operations are implemented through window scanning.
Answer:D
355, Which of the following Atlas products are used in training scenarios?
A. Atlas 300 (model 9000)
B. Atlas 500
C. Atlas 800 (model 9000)
D. Atlas 900
Answer:A
356, Which of the following are common tensor operations in Mind Spore?
A. asnumpy ()
B. sized ()
C. switch ()
Answer:A
357. Which two of the following are included in Mind Spore?

B. Mind Expression
C. Graph Engine
Answer:A,C,D
172
358. HUAWEI CLOUD Model Arts is 4 one-stop Al development platform. Which of the following functions can be provided
by Model Arts?
A. Data governance
B. Automatic learning
C. Visualized workflows
D. Al market
Answer:B
359. What are the major application scenarios of ModelArts?

A. Image
B. Speech
C. Text
D. Industrial control
Answer:A
360. What are challenges in Al capability development and application?

A. High barrier to entry
B. Low efficiency
C. Diverse requests
D. High computing power
Answer:A

A. True
B. False
Answer: B
362. The bottom layer of the random forest algorithm of an integration model is the ID3 tree.
A. True
B. False
Answer: A
363. In polynomial regression, a square term exists in a model's formula. Therefore, the model is nonlinear.
A. True
B. False
Answer: A
364. At present, Al is still in the narrow Al phase.

A. True
B. False
Answer: A
365. Computer vision is a science of how to make computers “see".

A. True
B. False
Answer: A
366. TensorFlow is now one of the most popular deep learning frameworks
A. True
B. False
Answer: A
367. Porch & 4 Python-based machine Rarning computing framework released by Facebook. It is developed bated on
Torch, 4 scientific computing framework supported by a large number of machine learning algorithms.
A. True
B. False
Answer: A
173
A. True
B. False
Answer: A
True
False
Answer: B
A. True
B. False
Answer: A
371. The CPU improves Al performance by adding instructions and cores

A. True
B. False
Answer: A
372. The typical application development process of the application development service includes mode! Evaluation.
A. True
B. False
Answer: A
373. Mind Spore supports interconnection with third-party chips and helps developers increase Mind Spore application
scenarios and expand the Al ecosystem.
A. True
B. False
Answer: A
A. True
B. False
Answer: A
375. The HIAI mobile computing platform provides various upper-layer function service APIs to ensure efficient running on
mobile devices.
A. True
B. False
Answer: B
376, In machine learning, () needs to be input to a model to train the model and make predictions.
A. Artificial programs
B. Historical data
C. Neural networks
D. Training algorithms
ANSWER : B
377. For multiclass problems, softmax assigns probability values to each class. The sum of these probability values is ( ).
A . 0.1
B. 0.5
C. 1
D . 100
ANSWER : C
174
378. What is the most important difference among batch gradient descent, mint-batch gradient descent, and stochastic
gradient descent?
A. Learning rate
B Gradient direction
C Batch size
D Gradient size
ANSWER : A
379. The three phases of Al include computing intelligence, ( ), and cognitive intelligence.
A. Behavior intelligence
C. Narrow Al
D. Strong Al
ANSWER : A,B
380, What is the purpose of adding the dropout layer when creating a network layer?
C. Preventing overfitting
ANSWER : C
381. What is the function of keras.preprocessing of TensorFlow 2.0?

B. Keras model generation tool
C. Keras model deployment tool
ANSWER : A
A. dimens
B. dtype
C. device
D. ndim
ANSWER : D
383. What is the meaning of the epochs parameter in model.fit(mnist train image, mnist.train.labels.epochs=5)?
B. The whole training set will be divided into six parts.
ANSWER : C

175
A. Google
B. Huawei
C. Baidu
D. Facebook
ANSWER :D
385. Which of the following is the output range of the ReLU function?
A. [0,+2)
B. (0,1)
C [1,1]
D. [-1,0]
ANSWER : D
386 .Which of the following statements about backpropagation is incorrect?

B. Backpropagation is used to update the weights in neural network .
D. Backpropagation can collaborate with a gradient descent algorithm to update a network weight
ANSWER : D
.
387. each layer of a neural network uses the Sigmoid activation function and the number of layers is large, which of the
A. Gradient explosion protein
B Vanishing gradient pattern
ANSWER : B
388. Which of the following statements about the feedforward neural network is correct?
A. An Input node of the feedforward neural network has a computing capability.
B. Neurons at the same layer are interconnected.
C. Inter-layer information is transmitted in only one direction.
D. Neurons with computing capabilities connect the upper and lower layers
ANSWER :
.
389. Which of the following statements about a recurrent neural network is incorrect?
A. The short form of recurrent neural network is RNN.
B. A recurrent neural network can be expanded according to the time axis.
C. Long short-term memory (LSTM) is a recurrent neural network.
D. GRU is not a recurrent neural network.
ANSWER : C
390, In 1958, Rosentsatt invented the perceptron algorithm, which was used for about 10 years. In 1969, Minsky
questioned the capability of the perceptron to resolve nonlinear problems and proposed the farnous problern that slowed
down the development of perceptron. The problem is an
A . AND problem
B. OR problem
C. XOR problem
D. XAND problem
ANSWER : D
176
A. dimens
B. dtype
C. device
D. ndim
392, What is the meaning of the epochs parameter in model.fit(mnist train image, mnist.train.labels.epochs=5)?
B. The whole training set will be divided into six parts.

A. Google
B. Huawei
C. Baidu
D. Facebook
395. Which of the following is the output range of the ReLU function?
A. [0,+2)
B. (0,1)
C [1,1]
D. [-1,0]
396 .Which of the following statements about backpropagation is incorrect?

B. Backpropagation is used to update the weights in neural network .
D. Backpropagation can collaborate with a gradient descent algorithm to update a network weight
.
397. each layer of a neural network uses the Sigmoid activation function and the number of layers is large, which of the
A. Gradient explosion protein
B Vanishing gradient pattern
398. Which of the following statements about the feedforward neural network is correct?
A. An Input node of the feedforward neural network has a computing capability.
B. Neurons at the same layer are interconnected.
C. Inter-layer information is transmitted in only one direction.
D. Neurons with computing capabilities connect the upper and lower layers
.
399. Which of the following statements about a recurrent neural network is incorrect?
A. The short form of recurrent neural network is RNN.
B. A recurrent neural network can be expanded according to the time axis.
C. Long short-term memory (LSTM) is a recurrent neural network.
D. GRU is not a recurrent neural network.
400, In 1958, Rosentsatt invented the perceptron algorithm, which was used for about 10 years. In 1969, Minsky
questioned the capability of the perceptron to resolve nonlinear problems and proposed the farnous problern that slowed
down the development of perceptron. The problem is an
A . AND problem
B. OR problem
C. OR problem
D. XAND problem
401, How many gates are there in GRU?

177
A1
B.2
C.3
D4
402. Features are dimensions that describe sample characteristics Which of the following ts a Correct statement about the
interpretability of features in traditional machine learning and deep learning?
A. Features have high interpretability in both traditional machine leaning and deop learning.
B. Features have low interpretability in both traditional machine learning and deep warming,
C. Features have high interpretability in traditional machine teaming but low interpretability in deep
learning..
D. Features have low interpretability in traditional machine teaming but high interpretability in deep
learning.
403, Which of the following optimizers can automatically adjust the learning rate?
A. SGD algorithm
404. Ina neural network, which of the following methods is used to update parameters during network training to minimize a
boss function?
B.Backpropagation algorithm
C.Convolution calculation
D. Pooling computing
405. Which of the following statements is Ascend Al Computing Engine?

A. AICPU
B. DvpP
C. Da Vinci Architecture
D. Buffers
406, How many layers does the software stack of an Ascend Al chip consist of?
A3
B.4
C. 2
D .5
407, Cell provides basic modules for defining and executing computing. Cell objects can be directly executed. Which of the
following statements is incorrect?
A. _ init, initializes and verifies components such as Parameter, Cell, and Primitive.
B .construct defines the execution process. In graph mode, a graph is compiled for execution, without
syntax restriction,
C. bprop (optional indicates the reverse direction of customized modules.
D. There ate about some predefined cells such as common optimizers for optim and common network
packaging functions for wrap,
408, Which of the following processing modes is used by MindSpore to implement efficient distributed training for ultra-
large models?
A. Auto parallel
B. Manual paratie!
C . Serial
409. Which of the following is not the application of the image recognition service?
A. Scenario analysis
178
B. Smart album
C. Object detection
D. Speech synthesis
410. Which of the following modules provides face detection on the HUAWEI HiAl platform?
A. HiAl Framework
B. HiAl Service
C. HiAl Foundation
D. HiAl Engine

B. SVM
D. Decision tree
A. Data acquisition
C. Model building
413, Which of the following options are the advantages of feature selection?
414, The F1 value for evaluating classification models is the harmonic mean of indicators including ().
A. Accuracy
B. Precision
C. Recall
D. Validity
415. Which of the following may be the segmentation surface of the support vector machine (SVM) machine learning
algorithm?
A. Straight line
B. Plane
C. Hyperplane
D Surface
A. Rules are complex or cannot be described, such as facial recognition and speech recognition.
B. The rules of the task will be changed with time, such as part-of-speech tagging, new words or
the trend of commodity sales.
D.. Problems have low complexity and a small scale.
417. Which of the following are Al application fields?

A. Smart home
B. Smart healthcare
C. Smart city
418. Which of the following libraries are used to process images and visualization toolkits?
A. Matplotlib
B. scikit-image
C. Opencv
179
D. scikit learn
419. Which of the following methods can be used to check whether a data type is tensor in TensorFlow 2,07
A, Is tensor
B. bsinstance
C device
D. dtype
420, Which of the following are built-in optimizers of the Keras module in TensorFlow?
A. Adam
B. AdaDelta
C. RMSprop
D. AdaBoost
421, Which of the following methods can be used to resolve data imbalance problems in deep learning tasks?
A. Random undersampling
D. Batch deletion
422. In convolutional neural networks, which of the following scenarios are suitable for using pooling layers?
A. Flower recognition
B. AlphaGo
C. Medical image segmentation
D. Face recognition
423. A generative adversarial network is like a game system A generator generates forged samples, and @ discriminator
determines whether the samples are true of fase
What is the ideal result?
A . The discriminator cannot determine whether the samples generated by the generator are true or tase
B.The discriminator can efficiently determine whether the samples generated by the generator are true or false.
C. The generator produces 4 large number of similar samples,
D. The distribution of generated surges fits the distribution of real samples
.
424, Which of the following statements about a convolutional neural network pooling layer are correct?
A. The pooling layer provides the function of dimension reduction.
B. Max-pooling and average-pooling are common pooling methods.
C. Featured images become smaller after pooling.
D. Pooling operations are implemented through window scanning.
425, Which of the following computing resources are included in the Da Vinci architecture?
A. Cube unit
B. Vector unit
C. Scalar unit
D. Tensor unit
426, Which of the following are Al training and inference frameworks?

A. TensorFlow
B. PyTorch
C. MindSpore
D. Pandas
427, Which of the following description about the modules of MindSpore is true?
A. model_zoo:Defines common network models
B. communication:Data loading module, which defines the dataloader and dataset and processes data
such as images and texts.
C. context Defines the context class and sets mode! running parameters, such as graph and Py Native
switching modes,
180
D. nn:Defines MindSpore cetis (neural network units), loss functions, and optimizers.
428, What are the major application scenarios of ModelArts?

A. Image
B. Speech
C. Text
D. Industrial control
429. Which of the following are features of the intrusion detection service provided by HUAWEI CLOUD El?
A. High flexibility
B. Ease-of-use
C. Low misreporting rate
D. High labor cost
430, What are the advantages of the HiAl mobile computing platform?
A. Comprehensive documents
B. Source code enabling quick start
C. Different types of APIs
D. Various tool chains

Hcia Ai Dump File

Uploaded by

Copyright:

Available Formats

Hcia Ai Dump File

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hcia Ai Dump File

Uploaded by

Copyright:

Available Formats

1 Dump Detail

1. Page 2 to 91 ============== Dumps

C. Image style migration

2. Which of the following about the gradient descent is incorrect?

A. Random gradient descent is a commonly used one in gradient descent.

C. The gradient descent algorithm 1s fast and reliable [Right Answers}

D. Large-scale distributed cluster

4. From the perspective of technical architecture Al The classification of chips includes?

6. TensorFlow in keras The built-in optimizer of the module has?

in deep learning is:

A. As the output value of the generated model

B. As the input value of the discriminant model

C. As the output value of the discriminant model

D. As input value for generative model

B. OBS Object storage service

D. El Big data service

A. L1 Regularization can do feature selection

B. L1 with L2 Regularization can be used for feature selection

C. L2 Regularization can do feature selection

D. L1 with L2 Regularization cannot be used for feature selection

B. Batch gradient algorithm can solve the local minimum problem

D. The convergence process of the global gradient algorithm is time-consuming

can provide parallel computing capabilities.:

22. What are the advantages of Python? (Multiple choice)

23. Pytorch Which company launched it first?

A. Any Python statement

B. while loop statement

C. for loop statement

D. Nested loop statement

26. The following description of machine learning is correct?

A. Deep learning is a branch of machine learning

B. Deep learning and machine learning are mutually inclusive

D. None of the above

28. What is wrong description of the normal distribution?

B. The normal distribution lakes the maximum value at the mean

D. Natural language processing

32. tf.keras.datasets Can view keras The built-in data set.

C. Discrete random variables have no d1slribulion function

D. Distribution function describes the value rule of random variables

B. The parameters need to be placed in parentheses

C. The function content starts with a colon and needs lo be indented

D. Return the result with return and the function ends.

scenarios covered? (Multiple Choice)

A. Financial product recommendation

C. Telecom customer retention

A. GBDT Algorithm is easier to underfit than random forest

B. Random forest is calculated in parallel, while GBDT Can't

C. GBDT Algorithm is easier to overfit than random forest

41. What is wrong about the image content review service?

A. politics Test results for sensitive persons involved in politics

B. terrorism Test results for political violence

C. confidence Represents confidence, range 0-100

D. labe1 Label representing each test result

A. Each key and its corresponding value need to be separated by ":

B. Separate the different key-value with ,

C. The entire dictionary is included in the {}

A. The number of rows is greater than the number of columns

B. The number of rows is equal to the number of columns

C. The number of rows is less than the number of columns

D. The number of rows has no relationship with the number of columns.