Improve Your Knowledge with H13-321_V2.0-ENU Exam Dumps

Category:

Comments:

0 Comments

Post Date:

August 28, 2024

Practicing with H13-321_V2.0-ENU questions can help you identify areas where you need to improve your knowledge. By answering H13-321_V2.0-ENU questions and reviewing your responses, you can identify gaps in your understanding and focus your study efforts on those areas. The H13-321_V2.0-ENU exam has a strict time limit, and you need to manage your time effectively to answer all the questions. Practicing with Huawei H13-321_V2.0-ENU dumps questions can help you develop time management skills by simulating the exam's time constraints. You'll learn how to pace yourself, manage your time effectively, and ensure that you complete the H13-321_V2.0-ENU exam within the allotted time. Test Huawei H13-321_V2.0-ENU exam free dumps below.

Page 1 of 18

1. Which of the following statements about tanh is correct? (Multiple choice)

Tanh also has the same disadvantage as the Sigmoid function, that is, the position derivative away from the center tends to 0

Tanh is symmetric about point 0, and the mean of the output is closer to 0 than Sigmoid, so SGD will be closer to the natural gradient, thereby reducing the number of iterations required

Tanh function can be seen as a logistic function that amplifies the translation

Tanh value range is (0, 1)

2. What are the evaluation indicators used in image segmentation?

3. Automatic learning is a service provided by ModelArts. It can automatically design models, automatically adjust parameters, automatically train, automatically compress and deploy models based on labeled data. It does not require code writing and model development experience, and can build AI models from scratch.

True

False

4. The deeper the deep neural network, the better the effect and the higher the accuracy.

True

False

5. In the image preprocessing histogram experiment, the cv2. equalizeHist (im) function represents the call to the OpenCV histogram equalization API.

True

False

6. Which of the following is the API used in mox.file to copy a file?

mox.file .copy

mox.file .rename

mox.file .copyparallel

mox.file .remove

7. Which of the following is not a NLP language feature?

Contextual word features

Text Features

Texture features

Single word features

8. What are the types of pooling? (Multiple choice)

Max Pooling

Average Pooling

Minimum Pooling

Convolution Pooling

9. What is Gaussian distribution also called?

Random distribution

Discrete distribution

Normal distribution

Abnormal distribution

10. In Huawei's artificial intelligence product stack, it can be compared to NVIDIA 's CUDA and CUDNN ()

Page 2 of 18

11. The HMM algorithm is a generative model, and CRF is a discriminative model. Both can be used for sequential labeling.

True

False

12. Images with low grayscale levels have richer layers and can show more details than images with high grayscale levels.

True

False

13. Which of the following operators can be used to implement edge detection through the filter method? (Multiple choice)

Laplacian operator

Sobel operator

Prewitt operator

Gaussian operator

14. Which of the following is an inference server provided by Shengteng Computing to its customers and partners?

Atlas 800 model

Atlas 300 model

Atlas 800 model

15. Digital images can perform basic mathematical operations, that is, addition, subtraction, multiplication and division.

True

False

16. Use two 3x3 convolution kernels to convolve a three-channel color image. How many channels does the resulting feature map have?

17. After successfully calling the speech recognition service, in the returned information, the ________ field contains the recognition result

18. AI approaches or even surpasses human capabilities in areas such as image recognition, speech recognition, game decision-making, and reading comprehension.

True

False

19. When calling Huawei cloud computing voice service, you can call the underlying interface encapsulated by SDK, and you can access the Restful service by simulating the front-end browser. The authentication methods used by the two are?

AK/SK, Token

Token, AK/SK

Token, Token

AK/SK, AK/SK

20. Which of the following options are the main advantages of the Keras interface in TensorFlow 2.0? (Multiple choice)

User-friendly

Modularity and composability

Easy to expand

Static graph mechanism is used

Page 3 of 18

21. ELU increases the computational complexity a little bit based on ReLU , but it has continuous derivatives and the defined surface is relatively smooth.

True

False

22. Which of the following indicators cannot be output by the classification_report in the sklearn.metrics model evaluation indicator module?

f1-Score

recall

accuracy

precision

23. Use OpenCV to directly mirror the image. Complete the following horizontal mirror code: im_flip = cv2.flip(im, ())

24. Which of the following templates can achieve image sharpening?

0 -1 0 -1 4 -1 0 -1 0

0 1 0 1 3 1 0 1 0

0 -1 0 -1 6 -1

0 1 0 1 1 1 0 1 0

25. Syntactic analysis generally uses a syntax tree to represent the analysis results.

True

False

26. Regarding Gaussian mixture model, which of the following statements is correct? (Multiple choice)

Gaussian mixture model is a linear combination of multiple single Gaussian models

Gaussian mixture model can use EM algorithm to learn model parameters

The number of sub-models K of the Gaussian mixture model needs to be given in advance

Gaussian mixture model can theoretically simulate any form of distribution

27. In order to solve the gradient disappearance problem, GRU uses () operation at the hidden layer output, t-1 state and t hidden layer state

28. Regarding CBOW and Skip-Gram algorithms, which of the following statements is incorrect?

Both CBOW and Skip-Gram can be used to train word embeddings, but Skip-Gram is faster than CBO

CBOW calculates the probability of a word appearing based on the n words preceding it or the n consecutive words before and after it.

Skip-Gram calculates the probabilities of several words before and after a certain word.

Both the CBOW model and the skip-gram model are based on the Huffman tree.

29. When calling the face comparison service, the person's gender is also one of the return values.

True

False

30. How to label object detection tasks in data management?

Use a circle to frame the target

Use a rectangle to frame the target

Use an ellipse to frame the target

Enter a category label for the image

Page 4 of 18

31. Power of Shengteng 310: ()

32. Which of the following options is the type of classification task label?

Discrete

Continuous

Automorphic

Strain type

33. In the field of deep learning, which of the following options are correct? (Multiple choice)

Update all parameter values through learning rate and gradient value to increase the loss value of the network

Gradient descent is actually to find the partial derivative (gradient) of the loss value with respect to all parameters. The meaning of the gradient is the direction of the fastest change.

Update all parameter values through learning rate and gradient value to reduce the loss value of the network

Learning Rate (LR) is an important hyperparameter in gradient descent

34. Suppose you have searched for 10,000 Weibo data and have no other information. Based on this data, you want to create a spoof text classification model that can divide each text into three categories: positive, negative, and neutral.

Which of the following models can perform the above text classification problem? (Multiple choice)

SVM

TextCNN

LSTM

None of the above

35. The back propagation algorithm (BP algorithm) mainly has two stages: forward propagation stage and back propagation stage, which completes the parameter update work in the neural network.

True

False

36. What is the essence of ModelArts preset models?

Reinforcement Learning

Transfer Learning

Reinforcement Learning

GAN

37. Traditional machine learning models have limited model capacity. Therefore, when the feature dimension is large, the performance will be very poor. Therefore, we need to do feature extraction and selection in traditional machine learning projects.

True

False

38. The coordinates of the four diagonal neighborhood points corresponding to P(x, y) : (x+1, y+1), (x-1, y-1), (x-1, y+1), (x+1, y-1) The coordinates of the four 4-neighborhood points corresponding to P(x, y): ()

39. What are the methods for text vectorization? (Multiple choice)

word2vec

Doc2vec

str2vec

vec2vec

40. We often talk about the 720P video format. What does 720 represent?

Video frame rate

Image resolution height

Width of image resolution

Grayscale of the video

Page 5 of 18

41. The English abbreviation of speech synthesis:

42. Which of the following is not a common method for text vectorization?

DBOW

CBOW

43. When doing character recognition on a scanned document, because the scanned document is not placed properly, the text is tilted at an angle, making it difficult to recognize. Which of the following technical means can be used to correct it? (Multiple choice)

Rotation transformation

Affine transformation

Grayscale transformation

Perspective Transformation

44. In the image preprocessing filtering experiment, which of the following functions calls OpenCV's median blur API?

cv2.GaussianBlur(im, (5, 5), 0)

cv2.blur(im, (3, 3))

cv2.medianBlur(im, 5)

cv2.filter2D(im, -1, sharpen_1)

45. Which of the following statements about participles is incorrect?

Rule-based word segmentation is simple and efficient, but dictionary maintenance is difficult.

The purpose of statistical segmentation is to perform probability calculation on the segmentation results and obtain the segmentation method with the highest probability.

Chinese is different from English natural word segmentation. Chinese word segmentation is a basic step in text processing. The performance of word segmentation directly affects the performance of other modules such as part of speech and syntax tree.

In actual engineering applications, word segmentation generally only uses one method.

46. Which of the following is not an advantage of the ReLU series of activation functions over the Sigmoid and Tanh activation functions?

The unilateral inhibition of ReLU provides the sparse expression ability of the network.

The non-saturation of ReLU can effectively solve the problem of gradient disappearance and provide a relatively wide activation boundary.

ReLU can cause neuron death during its training process.

From a computational perspective, both Sigmoid and Tanh activation functions require exponential calculations and are highly complex, while ReLU only requires a threshold to obtain an activation value.

47. The back propagation algorithm is an important method for updating parameters of neural networks. The back propagation algorithm is implemented with the help of which of the following calculation methods?

Computational Graph

Higher-order differentials

Cost Function

Chain Rule

48. Which of the following scenarios can text recognition technology be applied to? (Multiple choice)

Posture recognition (generally refers to human posture)

Invoice Identification

ID card recognition

Expression recognition (face recognition)

49. Which of the following scenarios involve speech recognition or speech synthesis technology? (Multiple choice)

Intelligent Customer Service

Intelligent distance education

Smart Home

Mobile voice assistant

50. In the gensim package, the word2vec parameter sg=1 means selecting the CBOM algorithm.

True

False

Page 6 of 18

51. All color images in computers are represented by RGB color space.

True

False

52. From which channels can image classification data for automatic learning be obtained? (Multiple choice)

Import from ModelArts Marketplace

Upload from the web page

Upload from OBS

Load from other cloud service providers' storage services

53. Which of the following belongs to convolutional neural network? (Multiple choice)

RNN

LSTM

VGG16

Resnet

54. Which of the following descriptions about probability and likelihood is correct? (Multiple choice)

If the random variable X follows a certain distribution, the likelihood refers to the authenticity of a set of parameters under the condition that X=x.

If the random variable X follows a certain distribution, the probability refers to the possibility that X=x under given parameters.

If the random variable X follows a certain distribution, the likelihood refers to the possibility that X=x under given parameters.

If the random variable X follows a certain distribution, the probability refers to the authenticity of a set of parameters under the condition that X=x.

55. Which of the following applications belong to supervised learning tasks in machine learning? (Multiple choice)

Expert System

Handwritten Character Recognition

Alpha

Image label classification

56. Among the following options, which of the following is the correct statement about AK/SK? (Multiple choice)

When sending requests to lower-level services through the API gateway, AK and SK must be used for signing.

Use the access key ID and the secret access key together to encrypt and sign the request

SK, a key used in conjunction with the access key ID to cryptographically sign the request, identifies the sender, and prevents the request from being modified

AK, Access Key ID, a unique identifier associated with the Secret Access Key

57. Which of the following options does not belong to the understanding and analysis of natural language?

Structural Analysis

Syntactic Analysis

Lexical Analysis

Voice Analysis

58. When paying by face recognition, which of the following steps is correct?

Image preprocessing - Image recognition - Face detection

Face recognition - image preprocessing - image detection

Image preprocessing - object detection - face recognition

Image preprocessing - face recognition - face segmentation

59. Match the following functions with the corresponding service modules

60. Which of the following options is Huawei’s artificial intelligence computing platform?

ModelArts

Huawei Cloud EI

ECS

Atlas

Page 7 of 18

61. The semantic content of an image may be very rich. Image content detection can understand the image content more intelligently and accurately. It can generate multiple description tags based on the image content. Which scene can it be applied to? (Multiple choice)

Photo retrieval and classification (generated description information can be used for image classification)

Restoration of Blurred Images

Targeted recommendation of image ads

Smart album management

62. The back propagation algorithm can be programmed to calculate the gradient of the neural network weights, thereby completing the training of the neural network through gradient descent.

True

False

63. M ultilayer perceptron can solve the XOR problem.

True

False

64. Which of the following processing flows belong to image preprocessing? (Multiple choice)

Object Detection

Geometric Correction

Motion Tracking

Histogram Normalization

65. Many mobile phones provide eye protection screen display mode, which can reduce blue light and relieve fatigue. This function can be achieved by adjusting the image ()

Brightness

Saturation

Contrast

Hue

66. Which of the following is not a speech acoustic feature?

Duration

Semantics

Frequency

Amplitude

67. ModelArts model conversion function can be used to convert and quantize Tensorflow models into tensorRT format. The converted model can be deployed and run on Nvidia p4 CPU.

True

False

68. Modelarts' image labeling processing is charged based on time.

True

False

69. A histogram is a graph with grayscale as the horizontal axis and grayscale frequency as the vertical axis.

True

False

70. Which of the following options is not a commonly used algorithm for keyword extraction?

TextRank

SSA

LDA

TF-IDF

Page 8 of 18

71. How to label image classification tasks?

Use a circle to frame the target

Use an ellipse to frame the target

Use a rectangle to frame the target

Enter a category label for the image

72. With the initial probability and transition probability matrix, the probability of any state can be calculated.

True

False

73. Huawei Cloud EI provides a Restful API. Which of the following options is a method for requesting the server to add new resources or perform special operations?

GET

POST

PUT

Attach

74. What kind of applications can the predictive analysis function in ModelArts automatic learning be used for?

Predict weather temperature changes

Identify flower species

Segmenting objects in images

Face recognition

75. Which of the following is a classification performance measurement indicator? (Multiple choice)

Accuracy

False rate

recall

precision

76. Which of the following operations can be used for image data enhancement? (Multiple choice)

Affine transformation

Channel Changes

Contrast Adjustment

Cut

77. Which of the following is a speech processing task? (Multiple choice)

Language Production

Language Comprehension

Speech Synthesis

Speech Recognition

78. Which of the following belongs to the sequence labeling task? (Multiple choice)

Sentiment Analysis

Chinese word segmentation

Entity Recognition

Part-of-speech tagging

79. The waveform sequence of an audio file is the information of the signal in the time domain.

True

False

80. In an 8-bit RGB color image, what color does (255, 255, 255) represent?

Blue

Red

Black

White

Page 9 of 18

81. The expressive power of a single perceptron is limited. It can only express linear decision surfaces (hyperplanes). If we interconnect many perceptrons, just like the human brain, we can express a wide variety of nonlinear surfaces.

True

False

82. What is the purpose of using gradient descent?

Finding label values

Changing the model structure

Optimization parameters

Prediction results (forward calculation in neural network)

83. Which of the following Huawei Cloud EI natural language processing services provide basic and domain APIs? (Multiple choice)

Text Classification

Sentiment Analysis

Text Summarization

Named Entity Recognition

84. Abbreviation of Bidirectional LSTM:

85. The Markov characteristic in the HMM model is reflected in the fact that the occurrence of the state at the next time point is related to all historical states.

True

False

86. When doing speech recognition tasks, use the Seq2Seq model. The input is a () sequence and the output is a letter sequence.

87. Which of the following belongs to the LDA training process? (Multiple choice)

Initialization

Gibbs sampling

Generate statements

Generate topics

88. Which of the following descriptions of the bimodal adaptive threshold segmentation method are correct? (Multiple choice)

The bimodal method divides the image into foreground and background according to the grayscale characteristics of the image. The larger the inter-class variance between the foreground and the background, the greater the difference between the two parts that make up the image.

If the image grayscale histogram is obviously bimodal, the grayscale value corresponding to the valley between the two peaks is selected as the threshold for image segmentation.

Histogram bimodal method is an easy-to-understand adaptive threshold algorithm

If the grayscale histogram of the image is very flat or has multiple peaks, the bimodal method cannot be used to obtain a suitable threshold.

89. What tasks in the speech field are suitable for GMM? (Multiple choice)

Speech Noise Reduction

Noise Tracking

Speech Synthesis

Speech Recognition

90. Natural language is a symbolic system with speech as its material shell, consisting of vocabulary and grammar. It is a natural language like Java, C++ and other languages..

True

False

Page 10 of 18

91. What are the time domain parameters of speech signals? (Multiple choice)

Short-term energy

Short-term zero-crossing rate

Short-term autocorrelation coefficient

Short-time average amplitude difference function

92. A photo has many small noise points during storage. The best denoising effect is to perform () operation on the scanned copy.

Median Filter

Gaussian filtering

Mean Filter

Laplace filtering

93. The Gaussian model uses the Gaussian density function to accurately quantify things and decomposes one thing into several models based on Bernoulli distribution.

True

False

94. HMM is composed of two random phenomena, one is from hidden state to hidden state, and the other is from hidden state to visible state.

True

False

95. Which activation function is used by the forget gate in LSTM?

Sigmoid

Tanh

ReLu

Softmax

96. Word vector is also called ()

97. Natural language processing is a hot field at present, but there are many problems and difficulties in its engineering process. What are the difficulties in named entity recognition? (Multiple choice)

The nesting situation is complex.

The length is uncertain.

The composition rules of named entities are complex.

There are a large number of named entities of various types.

98. The hidden layer is the intermediate layer between the input and output layers in the neural network. The training data does not give the output results required by these hidden layers.

True

False

99. Which of the following is not a field supported by the intent understanding API provided by Huawei Cloud EI natural language processing service?

Music

Alarm clock

Jokes

Entertainment

100. In Huawei Cloud Face Recognition Service, users can create multiple face libraries without restrictions.

True

False

Page 11 of 18

101. What are the tasks of machine learning scenarios? (Multiple choice)

Regression

Classification

Clustering

Generate

102. LSTM does not belong to the RNN network.

True

False

103. Markov chain refers to a discrete event random process with Markov properties in mathematics.

True

False

104. Which of the following is the specific content of text analysis? (Multiple choice)

Text Normalization

Speech Synthesis

Speech Analysis

Prosody Analysis

105. The annotation work in data management support third-party annotation.

True

False

106. What is the dimension of the commonly used Mel-Frequency Cepstral Coefficients?

107. What are the basic steps of the EM algorithm? (Multiple choice)

Initialization parameters

E-step and M-step

Iterate until convergence

Solve equations

108. The output of most optical sensors is a continuous signal. The computer can only perform digital image processing after the continuous signal is converted into a discrete signal.

True

False

109. Which modules does ModelArts cover? (Multiple choice)

Data Management

Development Environment

AI Market

Model Management

deployment and launch

110. Named entity recognition refers to identifying entities with specific meanings in text. What are the main ones? (Multiple choice)

Place names

Time

Name

Organization Name

Page 12 of 18

111. GRU contains reset gate and ()

112. The bag-of - words model is the earliest text vectorization method that uses words as the basic processing unit.

Which of the following options is not a defect of the bag-of-words model?

Unable to preserve order information

Based on the distribution hypothesis

The curse of dimensionality

There is a semantic gap

113. What is the advantage of stochastic gradient descent compared to batch gradient descent?

Easier to converge (stochastic gradient descent convergence is more unstable)

Faster iteration speed

Higher accuracy

Stronger generalization ability

114. Huawei Cloud EI provides a RESTful API. Which of the following options is a method for requesting the server to return a specified resource?

GET

PUT

Attach

POST

115. What are the commonly used ways to connect pixels? (Multiple choice)

16 connections

12 Connectivity

4 Connectivity

8-connectivity

116. Which of the following scientists created the Lisp programming language?

Alan Turing

Kilfond Shannon

Marvin Minsky

John McCarthy

117. What are the preset models supported by ModelArts? (Multiple choice)

Image Classification

Image Detection

Gene sequencing

Image Segmentation

118. Which of the following is not a method of speech feature extraction?

PLP

RNN

MFCC

LPC

119. Images are stored in the form of matrices in computers.

True

False

120. GMM is a linear mixture of multiple Gaussian models.

True

False

Page 13 of 18

121. What are the jieba word segmentation modes? (Multiple choice)

Precision Mode

Full Mode

Search Engine Mode

Other modes

122. There are two main methods for semi-supervised learning: one is to perform unsupervised clustering first and then supervised learning. The other is Self-Learning.

True

False

123. mox.file.stat can be used to obtain the stat information of an OBS file or folder. Which of the following information is included in the stat information? (Multiple choice)

File size

If it a directory

Creation time

Creation timestamp

124. The acquisition process of digital images is similar to that of the human eye, using a large number of photosensitive sensors to form an array to acquire images. The quality of the image is determined by the number of sensor units, () and sensing performance.

125. Which of the following options are common unsupervised algorithms? (Multiple choice)

k-means

PCA

GMM

Spectral clustering

126. Please match the following functions with the corresponding service modules

127. How does the learning rate change when using the AdaGrad algorithm?

Reduce

Increase

Increase first and then decrease

Decrease first and then increase

128. () in CANN combines the operator fusion of the internal storage architecture of the Ascend chip to reduce operator calling overhead and memory movement, and improve performance.

129. During a training, the ReLU neuron cannot be activated on all training data. We call this the dead ReLu problem.

True

False

130. Which of the following are hidden units? (Multiple choice)

logisticsigmoid and hyperbolic tangent function

Rectified Linear Unit (another translation method of RELU)

Borel unit

maxout unit

Page 14 of 18

131. The gradient descent algorithm is used to update parameters and find the minimum point of the loss function. Its convergence point depends only on the cost function.

True

False

132. Set the reset gate of GRU to 1 and the update gate to ________, then you will get the standard RNN model.

133. Which of the following options is not a sequence annotation method?

CRF

MEMM

N-Gram

HNM

134. Which of the following options belong to the training servers provided by Shengteng Computing to customers and partners? (Multiple choice)

Atlas 800 model

Atlas 300 model

Atlas 800 model

135. The definition of "artificial intelligence" was first formally proposed in 1956. In the more than 60 years since then, artificial intelligence has experienced three low points in development, the so-called "winter", but its development has not stopped.

True

False

136. When calling Huawei Cloud services, which of the following information will be used for authentication? (Multiple choice)

Current time

Call service name

137. When using the image labeling API provided by Huawei Cloud, which of the following methods are supported for importing images? (Multiple choice)

Base64 encoding

Read directly from OBS

Huawei Cloud Disk

Url

138. Which of the following is correct about the Gaussian distribution curve?

The Gaussian curve is bell-shaped, with high ends and low in the middle.

The smaller the standard deviation, the flatter the curve

The Gaussian curve is bell-shaped, with low ends and high in the middle.

The larger the standard deviation, the taller the curve.

139. What are the preset object detection models supported by ModelArts? (Multiple choice)

Faster_RCNN_RestNet_v2_50

RestNet_v1_50

Inception_v3

Faster_RCNN_RestNet_v1_50

140. Use the RNN+CTC model for speech recognition. When generating the predicted output, the network predicts the corresponding letter sequence for the input audio feature sequence. The () algorithm can be used to search for the optimal path.

Page 15 of 18

141. Maximum likelihood estimation does not require specifying in advance that the data follows a certain distribution

True

False

142. Which of the following belongs to the content of linguistics? (Multiple choice)

Text

Voice

Vocabulary

Grammar

143. When ModelArts annotates a dataset, you can see the annotation progress, the number of annotated images, and the total number of images.

True

False

144. Which of the following options are related services of Huawei Cloud EI? (Multiple choice)

ModelArts

EI Agent

EI Experience Hall

EI service family

145. A function of tensorflow2.0 needs to display the results and information of intermediate training when outputting. Please write which parameter ________ in the function should be set to True.

146. What are the advantages of MoXing framework? (Multiple choice)

Automatic hyperparameter search

API usability packaging

Batch_size Adaptation

Distributed training acceleration

147. In the grayscale change experiment of image preprocessing, the gamma change function is defined as follows:

Xxxxxxxxxx

Def gamma_trans (img, gamma):

Gamma_list = [ np.power (x/255.0,gamma)*255.0 for x in range(256)]

Gamma_table = np.round ( np.array ( gamma_list )). astype (np.uint8)

Return cv2.LUT(img , gamma_table )

When the gamma value is 0.5, the bright part is stretched and the dark part is compressed.

True

False

148. Which of the following descriptions of the shortcomings of the co -occurrence matrix + SVD is wrong?

SVD only learns shallow semantics (correct)

The co-occurrence matrix does not take into account the order of words (correct)

SVD calculation is large

The co-occurrence matrix is high-dimensional and sparse (correct)

149. What are the DNN model acceleration methods? (Multiple choice)

Multi-GPU Backpropagation

Reduce the size of the model

Asynchronous Stochastic Gradient Descent

Scaling up the model

150. What types of image annotations does ModelArts support?

Image Classification

Object Detection

Image Segmentation

All of the above are true

Page 16 of 18

151. Deep learning model training requires a large amount of data. In actual operation, random noise can be added to the training image data set through data enhancement technology to expand the sample size of the data set.

True

False

152. What is the role of Gaussian mixture model in speech recognition?

Represents the probability transfer between factors

Multidimensional numerical distribution used to represent a phoneme

Used to distinguish different phonemes

Evaluating the Advantages of the HMM Algorithm

153. Which of the following factors can be used to describe the hidden Markov model? (Multiple choice)

Observation Set

Observation sequence

State sequence

State Collection

154. Which of the following descriptions of gradient disappearance and gradient explosion are correct? (Multiple choice)

The gradient explosion problem can be effectively solved by simple gradient clipping

Gradient explosion usually occurs in deep networks and when the weight initialization value is too large

One of the reasons for the disappearance of gradients is the use of inappropriate activation functions in deep networks, such as sigmoid

The gradient disappearance problem can be effectively solved by simple gradient clipping

155. What are the methods for extracting speech signal features? (Multiple choice)

Linear Prediction Coefficient (LPC)

Mel Frequency Cepstral Coefficient (MFCC)

Perceptual Linear Prediction (PLP)

Line Spectral Pair Parameter (LSP)

156. The architecture adopted by Huawei series chips is ()

157. () features and Adaboost algorithm are used together to achieve good results in face detection.

158. The (swish) function is obtained by (), so it does not need to be obtained through inferential calculation.

159. The denoising effect of mean filtering is better than that of Gaussian filtering for the same size.

True

False

160. A neural network can simulate any polynomial function as long as it has a hidden layer and an output layer and as long as the input data is sufficient.

True

False

Page 17 of 18

161. Which of the following is the abbreviation of speech synthesis?

TTS

ASR

STT

AST

162. The more parameters a deep neural network has, the better it is at capturing the characteristic information of the data.

True

False

163. The OTSU algorithm traverses possible thresholds and selects the maximum segmentation threshold between the foreground area and the background area () as the optimal threshold.

Absolute difference

Area difference

Cumulative difference of pixel values

Between-class variance

164. Huawei's full- stack, full-scenario AI solution refers to the chip enabling layer ()

165. Back propagation is a common method used to train artificial neural networks.

True

False

166. Use a 3*3 convolution kernel to convolve a three-channel color image. The number of parameters that need to be learned is?

167. What is the main function of MoXing? (Multiple choice)

Huawei's self-developed deep learning tool library

Huawei's self-developed deep learning engine

Distributed training acceleration framework for deep learning engines such as TensorFlow, encapsulation and acceleration of API interfaces

Simple encapsulation of deep learning engine API interfaces such as TensorFlow

168. Which of the following is the core idea of convolutional neural network? (Multiple choice)

Parameter Sharing

Zero padding

Fully connected

Local Perception

169. When using GMM-HMM for single word recognition, only one GMM model in a model needs to be trained.

True

False

170. Only when any technology is positioned accurately can its value be fully realized. Proper positioning of artificial intelligence technology is the basis for our understanding and application of this technology. Just like the wheel and iron in BC, the railway and electricity in the 19th century, and the car, computer, and Internet in the 20th century, artificial intelligence is a set of technologies and a new general-purpose technology.

True

False

Page 18 of 18

171. MoXing is a network model development API provided by Huawei Cloud ModelArts. Using the mox.file module in MoXing , you can directly access OBS by calling the API and download the folder from OBS to the local computer. Complete the code: mox.file .() ('s3:// my_bucket / input_data ', '/cache/ input_data ').

172. What are the common feature selection methods for text classification? (Multiple choice)

Chi-Square Test

Mutual Information

Information Gain

None of the above

TAGS:

H13-321_V2.0-ENU, H13-321_V2.0-ENU exam dumps

Notify of

Label

Name*

Email*

Website

Label

Name*

Email*

Website

0 Comments

Oldest

Newest Most Voted

Inline Feedbacks

View all comments

Get H13-321_V2.0-ENU Dumps Full Version

Q&As: 571
Versions: PDF and Software Download Now

About Dumpsinfo

Dumpsinfo is a good platform providing the latest exam information and dumps questions for all IT certification exams. You can study all the latest exam dumps questions online.

[email protected]

Mon - Sat 9:00am - 6:00pm

Improve Your Knowledge with H13-321_V2.0-ENU Exam Dumps

Related

Posts

H13-527_V5.0 Dumps Questions Increase Your Chance of Success

Get Certified Easily with Online H19-402_V1.0 Dumps

Prepare H13-611_V5.0-ENU Exam with Using H13-611_V5.0-ENU Dump Questions

Study Online H13-629_V3.0-ENU Exam Dumps to Pass

About Us

EMAIL

Services

Opening Hours