H13-323_V1.0 HCIP-AI Solution Architect V1.0 Exam Free Dumps

dumps2025-04-22T10:24:17+08:00

H13-323_V1.0 exam is one for HCIP-AI Solution Architect certification, which is popular among those Huawei candidates. Huawei H13-323_V1.0 exam basic information is available below for you to study. There are 60 questions in real Huawei H13-323_V1.0 exam, and you have 90 minutes to answer all the questions. The required passing score is 600. The exam cost is $300. We provide the latest HCIP-AI Solution Architect H13-323_V1.0 exam dumps, which can ensure you pass the test easily. Free HCIP-AI Solution Architect H13-323_V1.0 exam dumps are available below.

Page 1 of 11

1. Which of the following is correct?

The two objects generated by shallow copy and deep copy use == to judge that the result is True

The two deep-copied objects use is to check if the result is True

The shallow copy of the two objects uses is to judge the result is True

The result of using == to judge the two deep-copied objects is False

2. What is the HiAI resource package that HiAI partially releases to third-party developers called?

HiAIDDR

HIAI DDK

HIAI SDK

HIAIJDK

3. Which of the following schools of thought does neural network research belong to?

Symbolism

Connectionism

Behaviorism

None of the above

4. Which of the following is not a feature of MindSpore's core architecture?

Automatic Differentiation

Automatic tuning

Automatic encoding

Automatic parallelization

5. Singular value decomposition can be performed on non-square matrices.

True

False

6. ModelArts architecture includes AI data framework, algorithm development, training, and deployment.

True

False

7. Which of the following statements about the SVM algorithm is wrong?

The SVM algorithm can use a variety of kernel functions

BSVM can be used to solve classification problems

The support vector is the vector point closest to the hyperplane

The DSVM algorithm cannot be used to process nonlinear data sets

8. Current AI is still at the artificial intelligence level.

True

False

9. What methods can be used to merge tensors in TensorFlow 2.0?

Concat

split

join

unstack

10. When was TensorFlow 20 released? In which year?

2015

2017

2019

2018

Page 2 of 11

11. Where should labeled data be placed in a GAN?

As input value of the generative model

As input value of the discriminant model

As the output value of the generated model

As the output value of the discriminant model

12. Which of the following statements about convolutional neural networks is wrong?

Convolution kernels cannot be used to extract global features of images

When processing an image, the image is convolved in a scanning window manner.

Common pooling layers include maximum pooling and average pooling

Convolutional neural networks can contain convolutional layers, pooling layers and fully connected layers

13. Which module of HUAWEI HiAI does Huawei chip support?

HiAI Engine

HiAI Foundation

HiAI Framework

HiAI Service

14. Which of the following is an advantage of one-hot word embedding?

When the vocabulary is large, the word vector dimension will be too large

Simple and easy to understand

The words are isolated and cannot express similar meanings

Sparse expression of words

15. In a neural network based on connectionism, each fixed point can express a specific meaning.

True

False

16. TensorFlow is one of the most popular deep learning frameworks.

True

False

17. The fully connected layer is generally located in the middle of the network to reduce the number of parameters.

True

False

18. Which of the following statements about neural networks is wrong?

As the number of hidden layers of the neural network increases, the classification ability of the model gradually weakens

Feedforward neural networks can be represented by directed acyclic graphs

The limitation of a single-layer perceptron is that it cannot solve or

There is no connection between neurons in the same layer of the feedforward neural network

19. Federated learning uses different data sources to train models while ensuring data privacy and security, and further breaks through data bottlenecks.

True

False

20. All convolution kernels in the same convolution layer in a convolutional neural network share weights.

True

False

Page 3 of 11

21. Which of the following activation functions can solve the gradient vanishing problem well?

Tanh

Sigmoid

Softsign

Relu

22. In which year was the term “artificial intelligence” first published?

1956

1950

1959

1946

23. For a matrix with singular values, the matrix must be a square matrix and the singular values are arranged in descending order.

True

False

24. While creating a network layer, what is the main purpose of adding a dropout layer?

Extracting image features

Acts as a classifier

Increase nonlinearity

Prevent overfitting

25. Training a large model is a systematic project, covering multiple aspects including large model development, training, and fine-tuning. Improving the efficiency and scalability of large model distributed scale training is the main function of which of the following large model development kits?

Huggingface PEFT

DeepSpeed

MindFormers

MindPet

26. In machine learning, which of the following options is not part of the process from acquiring data to formally putting it into the model?

Data Standardization

Data cleaning

Data Dimensionality Reduction

Data Visualization

27. Concurrency refers to the execution of multiple tasks at the same time. Concurrent operations can be completed using multiple processes in Python.

True

False

28. HUAWEI HiAI Engine can easily integrate various AI capabilities with apps.

True

False

29. Regularization can also be added to logistic regression to avoid overfitting.

True

False

30. What data types does TensorFlow use?

Scalar

Matrix

Vector

Tensor

Page 4 of 11

31. Combining ModelArts service with () service can easily deploy models to the "end".

OBS

ORC

ECS

HiLens

32. Which of the following engines cannot be provided by HUAWEI HiAI Engine?

ASR Engine

NLU Engine

DSP Engine

CV Engine

33. Which of the following identifiers is incorrectly named?

a0*

_0a

34. Which tools can be integrated with the HUAWEI HiAI platform?

Jupyter Notebook

MyEclipse

Android Studio

Spider

35. Jobs that are repetitive and require weak social skills are easily replaced by AI.

True

False

36. When using the keras interface of TensorFlow20 to build a neural network, it is necessary to compile the network. Which of the following methods should be selected?

compile

write

join

fit

37. ModelSummary0 in Keras is used to print the network overview.

True

False

38. Which of the following options is correct about the relationship between Poisson distribution and binomial distribution?

The Poisson distribution has nothing to do with the binomial distribution.

Poisson distribution can replace binomial distribution

Poisson distribution is an approximation of binomial distribution when n is large and p is small

The mathematical models of Poisson distribution and binomial distribution are both Lagrangian probability models

39. Recurrent neural networks can capture dynamic information in serialized data.

True

False

40. The three main advantages of the Keras interface are user-friendliness, modularity and composability, and ease of extensibility.

True

False

Page 5 of 11

41. Which of the following statements correctly describes the synergistic effect when a large model is used in conjunction with a knowledge base?

Using large models in conjunction with knowledge bases limits the efficiency of information retrieval because large models process slower than traditional database queries.

By combining the large model with the knowledge base, it can quickly retrieve relevant knowledge and generate accurate answers, improving the efficiency and accuracy of information retrieval.

The combination of large models and knowledge bases is mainly used to enhance the gaming experience and provide richer game content and interaction.

Combining big models with knowledge bases can only be applied in the field of online education to provide customized learning content and tests

42. The loss function of logistic regression is the cross loss function.

True

False

43. A coin is tossed 100 times, 60 times it is heads and 40 times it is tails. What is the maximum possible distribution of the mass of the coin?

Back/front = 2:1

Back/front = 3:2

Back/front = 5:3

Back/front = 5:2

44. What does the epochs parameter in the code model.fit(mnist.train.inmage.mnist.train.labels.epochs=5) represent?

The entire training set will be trained 5 times

The entire test set will be tested 5 times

The entire training set will be divided into 6 parts

The entire training set will be divided into 5 parts

45. The covariance matrix is a symmetric matrix, and the elements on the diagonal are the covariances of pairwise random variables.

True

False

46. Which of the following statements about the combination of LangChain and RAG is wrong?

Data loader is used to load data from a data source and convert it into a document object.

The text segmenter divides the document object into smaller document objects to facilitate subsequent retrieval and generation.

Text embedder is used to convert texts into embeddings to measure the similarity between texts.

The memory does not require indexing technology and can directly store and query embeddings.

47. After the model is trained, it is generally necessary to compress and quantize the model before deploying it as a service and performing inference. Under the premise that the data meets the accuracy required by the model, the model's reference amount can be reduced. Which of the following tools is mainly used to compress large models and accelerate inference?

Mindspore

MindPet

MindFormers

ModelSlim

48. The Sys block is responsible for interacting with the operating system and provides an interface for accessing the underlying operating system.

True

False

49. Which of the following attributes does TensorFlow 2.0 not support in creating tensors?

create

f11

zeros

constant

50. Which of the following output results is wrong?

print(*map(lambda x: x*x, [1, 2, 3])) output: 149

print(*zip([1, 2, 3], ["a""], ["A""B""C")) Output: (1, a'"

(2, b', B(3, C

print(*filter(lambda x:x%2==1),[1,2,3])) output result 13

sorted([("b', 2), ("a, 1), (c', 3), (d', 4)], key=lambda x:x[1]) Output: [(a', 1), ("', 2), (c, 3), (d', 4)]

Page 6 of 11

51. Which of the following is the value range of INT8?

[-128, 128]

[0, 255]

[-127, 128

[-128, 127]

52. A first -rank tensor is a scalar.

True

False

53. PyTorch's current versions support Python2 and Python3.

True

False

54. Which of the following statements about gradient descent is correct?

Global gradient descent is faster than stochastic gradient descent

When using gradient descent, the learning rate should be kept consistent

Adam is not a gradient descent algorithm

Stochastic gradient descent is sensitive to noise points

55. Which company developed TensorFlow?

Oracle

Facebook

Google

NVIDIA

56. Which of the following options does not belong to Huawei's full-stack solution?

Application Enablement

Edge computing

Open Source Framework

Chip Enable

57. In the application of large language models, which technology can enhance the model's reasoning ability by simulating the human problem-solving thinking process?

Federated Learning

Reinforcement Learning

Thinking Chain

Transfer Learning

58. The Naive Bayes algorithm does not require independent and identical distribution between sample features

True

False

59. Which of the following statements about the standard RNN model is correct?

Back propagation does not consider the time direction

There is no one-to-one model structure

There will be a problem of information decay during long-term transmission of memory

There is no many-to-many model structure

60. Which statement about matrix is correct?

Non-square matrices also have inverse matrices

The transpose of an orthogonal matrix A is equal to the inverse of A

The transpose of a symmetric matrix A is A itself

Any matrix multiplied by the identity matrix is equal to itself

Page 7 of 11

61. Which of the following is not the core of ModelArts?

Cloud

Edge

End

API

62. The direction of the gradient is the direction of the maximum directional derivative.

True

False

63. What is the function of tf.keras.preprocessing?

Keras model deployment tool

Keras built-in optimizer

keras data processing tool

Keras Generative Model Tool

64. Which of the following does not belong to the development of large-scale chat robots?

More natural and smooth interaction experience

Richer expressive and generative capabilities

Shorter context processing capabilities

Better multimodal interaction capabilities

65. TensorFlow is an end-to-end open source platform for machine learning and deep learning.

True

False

66. Huawei's AI landscape includes public, IoT industry terminals, consumer terminals, and other end-to-edge, cloud deployment environments.

True

Flase

67. Which of the following statements about back propagation is wrong?

Backpropagation can only be used in feedforward neural networks

Back propagation will pass through the activation function

Back propagation can be combined with the gradient descent algorithm to update network weights

Back propagation refers to the error propagating backward through the network

68. An artificial neural network is an information processing system designed to mimic the structure and function of the human brain.

True

False

69. Which of the following statements about the seq2seq model is incorrect?

The Seq2seq model consists of two parts: encoder and decoder

Encoder and Decoder can be CNN, RNN, Transformer three structures

The Encoder and Decoder of Seq2Seq can only have the same structure

Encoder converts variable-length signal sequences into fixed-length vector expressions, which is essentially feature compression.

70. Which of the following options is not a function supported by the Megatron-LM parallel framework?

ZeRO Data Parallelism

Tensor Parallelism

Pipeline parallelism

BF16 optimizer

Page 8 of 11

71. Which of the following is the offline mode in HiAI?

HiAI Service

HiAI Engine

HiAI Foundation

HAIIDE

72. Python supports multiple versions coexistence.

True

False

73. What methods are not included in the built-in metrics of tfkerasmetrics?

mean

Accuracy

Recall

Sum

74. HiAI mobile computing platform provides a rich set of upper-layer functional business APIs, which can be efficiently run on mobile devices.

True

False

75. Tensor is the data storage board in MindSpore.

True

False

76. In TensorFlow2.0, gather is used to increase dimensions.

True

False

77. What is the label predicted by the algorithm?

Continuous

Strain type

Discrete

Automorphic

78. There may be a way to use double underscores in Python.

True

False

79. What are the signs of a person being at the perception stage?

Machines begin to understand, think and make decisions like humans

Machines begin to calculate and transmit information like humans

Machines begin to understand and hear, make judgments, and take some simple actions

80. On-Device execution, that is, whole-image offloading execution, fully utilizes the computing power of the Intel Xeon chips, can greatly reduce the interaction overhead, and thus increase the accelerator occupancy rate. Which of the following descriptions about On-Device execution is incorrect?

MindSpore achieves decentralized autonomous A11 Reduce through adaptive graph optimization driven by gradient data, with consistent gradient aggregation and full flow of computing and communication.

Challenges of model execution under super chip computing power: memory wall problem, interaction overhead, and difficulty in providing big data. Part of the model is executed on the host, and part of the model is executed on the device. The interaction overhead is even much greater than the execution overhead, resulting in low accelerator utilization.

MindSpore uses chip-oriented deep graph optimization technology to reduce synchronization waiting and maximize the parallelism of "data computing and communication". The training performance is better than the host side graph debugging method.

Challenges of distributed gradient aggregation under super chip computing power: When ResNet50 single-select generation takes 20mms, it will generate central control synchronization overhead and frequent synchronization communication overhead. The traditional method requires 3 synchronizations to complete A11Reduce, while the data-driven method has independent A11Reduce and no control overhead.

Page 9 of 11

81. What is the method to check the tensor dimension in TensorFlow20?

dimens

dtype

ndim

device

82. A matrix corresponds to a linear transformation, which is implemented through matrix multiplication.

True

False

83. Which of the following statements about supervised learning is correct?

Supervised learning is a rule-based algorithm

Decision tree is a supervised learning

Supervised learning does not require labels to train

Supervised learning cannot be trained using cross validation

84. In the text "Hello, I am Wang Ziwen, a teacher from Hangzhou", understanding Wang Ziwen as Wang Ziwen, a teacher from Hangzhou, belongs to which difficulty in natural language processing?

Pragmatic ambiguity

Semantic ambiguity

Syntactic ambiguity

Lexical ambiguity

85. Suppose we have a text fragment containing two sentences, namely: "Tony likes to eat apples" and "He likes eating apples". Now we use the pre-trained BERT model to encode this text fragment and then use the sentences "Tony" and "He" as queries respectively and calculate the similarity score of the two sentences through BERT.

How will the output vector of the last layer affect the calculation results?

The output vector of the last layer captures the meaning of the entire text segment and has a greater impact on the similarity score

The output vector of the last layer mainly captures the information of a single sentence and has limited impact on the similarity score

The output vector of the last layer will distinguish the keywords of the two sentences and have little impact on the similarity score

The last layer output vector has no effect on the similarity score because the similarity calculation only depends on the word level representation

86. Which of the following is not a commonly selected Operation by MindSpore?

signal

math

array

87. TensorFlow20 does not support the transfer of tensors between GPU and CPU.

True

False

88. The three stages of artificial intelligence include computational intelligence, (), and knowledge

Weak AI

Perceptual Intelligence

Behavioral Intelligence

Strong AI

89. Softmax function is often selected as the output layer of multi-classification tasks

True

False

90. A vector is just a number.

True

False

Page 10 of 11

91. The function of the activation function is to transform the linear mapping into a nonlinear mapping.

True

False

92. When a function is at a certain point in the variable space, along which direction does it have the largest rate of change?

Gradient

Minimum Directional Derivative

Directional derivative

Derivative

93. Which of the following statements about image content review services is incorrect?

politics is the detection result of politically sensitive figures

terrorism is a result related to political violence and terrorism

confidence represents the confidence level, ranging from 0 to 100

label represents each detection result label

94. The graph and session mechanisms were cancelled in TensorFlow20.

True

False

95. Which of the following options is not an evaluation indicator for parameter estimation?

Adaptive

Consistency

Unbiasedness

Validity

96. Which of the following descriptions about the Transformer model is incorrect?

Transformer consists of two parts: Encoder and Decoder

The layer structures in the Encoder are the same

Parameters are shared between layers in the Encoder

The sentence input from the encoder first passes through a self-attention layer

97. Which of the following is not a ModelArts development type?

Standard Model Development

Rapid Prototyping

Zero-based modeling

98. Computer vision research is the science of making computers "see"

True

False

99. Which of the following tables about L1 regularization and L2 regularization is correct?

After adding the L2 regularization term, the gradient descent algorithm cannot be used to iterate the parameter value

L1 and L2 regularization terms cannot act on the loss function

L2 regularization produces a more sparse model than L1 regularization

L1 regularization term helps to enhance the generalization ability of the model

100. Which of the following expressions can be used to represent 110 in the sentence “ My phone number is 110. Please contact me if you have any questions ” ?

re.split("d+", str1)

re.match("[0-9]+, str1)

re.search("d+", strl)

re.match("d+", str1)

Page 11 of 11

101. Single value decomposition can be performed on non-square matrices.

True

False

102. Which of the following statements about recurrent neural networks is wrong?

LSTM cannot solve the problem of gradient disappearance

LSTM is also a recurrent neural network

Recurrent neural network can be abbreviated as RNN

Recurrent neural networks can be expanded according to the time axis

103. What is the result of 5/2 in Python 3?

2.5

104. Which of the following statements about recurrent neural networks is wrong?

No action required

Dulicate OBS

Sync OBS

Move OBS

H13-323_V1.0 HCIP-AI Solution Architect V1.0 Exam Free Dumps

H13-323_V1.0 HCIP-AI Solution Architect V1.0 Exam Free Dumps

Share this post

Leave a Reply Cancel reply

Related Posts