Chapter test questions:
Chapter 1 Overview of Artificial Intelligence
1. What are the following options for the application direction of artificial intelligence?
A.Computer Vision
B. Natural language processing
C. Speech processing
D.Smart city
Correct answer: ABCD
2. The three main schools of artificial intelligence include:
A. Symbolism
B.Connectionism
C. Behaviorism
D. Activism
Correct answer: ABC
3. The current development of artificial intelligence belongs to the level of strong artificial intelligence.
Correct answer: False
4. Artificial intelligence, machine learning, and deep learning are mutually inclusive. Artificial intelligence includes machine learning, and machine learning includes deep learning.
Correct answer: Correct
Chapter 2 Overview of Machine Learning
1. According to certain specified attribute characteristics, it is divided into two or more categories. The questions it belongs to are:
A.Return
B.Association
C.Classification
Correct answer: C
2. Data cleaning includes:
A. Missing value processing
B. Outlier processing
C. Remove duplicates
D. Modify format errors
Correct answer: ABCD
3. Which of the following statements about ensemble learning are correct?
A. Integrated learning algorithms need to use decision trees as the underlying algorithm.
B. The voting strategy of ensemble learning can choose the averaging method.
C. When random forest trains subtrees, a batch of features are randomly selected for training.
D.Xgboost can be trained in parallel like Random Forest.
Correct answer: B.C.
4. (Judgment) K-Means can directly classify text.
Correct answer: False
5. Predicting the specific value of house prices based on existing house sales information is a regression problem.
Correct answer: Correct
6. The goal of clustering is to minimize the similarity between clusters and maximize the similarity within clusters.
Correct answer: Correct
Chapter 3 Overview of Deep Learning
1. Which of the following descriptions of the Softmax function is incorrect?
A. The function of the Softmax function is to map an arbitrary K-dimensional real number vector into another K-dimensional real number vector.
B. Each element in the Softmax function output vector has a value between [0, 1].
C. The sum of the module lengths of all dimensions of the output vector is not 1.
D.Softmax function is often used as the output layer for multi-classification tasks.
Correct answer: C
2. Which of the following is correct in the description of deep learning?
A. Use algorithms to automatically extract features.
B. Because of the parallel computing of GPU, the more network parameters, the faster the calculation speed.
C. Each layer parameter has a clear meaning.
D. Features are highly interpretable.
Correct answer: A
3. Which of the following descriptions of gradient descent are correct?
A. In SGD, because the samples selected for each training are random, this itself brings instability, which will cause the loss function to produce turbulence or even reverse displacement in the process of falling to the lowest point.
B.BGD training is the most unstable, but consumes too much computing resources.
C.MBGD is a method after balancing SGD and BGD, so this method is better for all data sets.
D. Batch Gradient Descent (BGD) uses all training samples for training each time.
Correct answer: ACD
4. Convolutional neural networks cannot be used to process text data.
Correct answer: False
5. Input a 32x32 image (no padding), use a convolution kernel of size 5x5 to perform convolution calculation with a step size of 2, and the output image size is 13*13.
Correct answer: False
6. In a convolutional neural network, the size of the convolution kernel and the size of the pooling layer window do not need to be consistent, but the step size must remain consistent.
Correct answer: False
7. Dropout will randomly discard a part of neurons during the training process. At this time, the corresponding parameters of the discarded part will not participate in the training in the next round of training.
Correct answer: False
Chapter 4 Artificial Intelligence Development Framework
1. Which of the following options is the basic unit of the network in MindSpore?
A. mindspore.nn.Dense
B. mindspore.nn.Cell
C. mindspore.nn.Base
D. mindspore.nn.NN
Correct answer: B
2. In MindSpore, which of the following is not a network operator?
A.Conv2D
B.SGD
C.ControlDepend
D. Softmax
Correct answer: C
3. Which of the following statements about MindSpore are correct?
A.MindSpore supports switching between dynamic and static images, and the API expression is consistent.
B. MindInsight is MindSpore's visual debugging tool, which can optimize model performance and debugging accuracy.
C.MindSpore provides dataset and mindrecord modules for data processing.
D.MindSpore supports data types such as Int, Uint and Float.
Correct answer: ABD
4. In MindSpore, you can use the decorator @ms_function to optimize the efficiency of existing dynamic and static images.
Correct answer: False
Chapter 5 Introduction to Huawei Artificial Intelligence Platform
1. Which of Huawei Cloud services is not part of the AI basic platform?
A.AI development platform ModelArts
B. Huawei HiLens
C. Cloud database RDS
D.Graph engine service GES
Correct answer: C
2. Which of the following products can Ascend 310 processor be applied to?
A.Atlas 200DK AI Developer Kit
B.Atlas 500 smart station
C.Atlas 200 AI acceleration module
D.Atlas 300 AI accelerator card
Correct answer: ABCD
3. What are the development modes supported by the training platform in ModelArts?
A. Cloud development (Codelab)
B. Cloud development (Notebook+SDK)
C. Local development (PyCharm+PyCharm ToolKit)
D. Local development (IDE+SDK)
Correct answer: ABCD
4. AICore, as the computing core of the Ascend Al chip, is responsible for complex matrix calculations.
Correct answer: Correct
Chapter 6 Frontier Application Scenarios of Artificial Intelligence
1. Which of the following options is not a key capability of Huawei’s autonomous driving cloud service?
A. Labeling platform
B. Model training
C. Parallel simulation
D. Sensor docking
Correct answer: D
2. In the knowledge map, the nature and relationship of a specific thing are called __, and this specific thing is called _.
A. Attributes, ontology
B. Concept, entity
C. Attributes, entities
D.Concept, attribute
Correct answer: C
3. Which of the following statements about generative adversarial networks is correct?
A. The discriminator needs to determine whether the image output from the generator is real.
B. The generator needs to generate fake graphics to "fool" the discriminator.
C. In order to ensure the realism of the generated images, a highly accurate discriminator is required when starting training.
D. During the training process, the discriminator will be trained first, then the discriminator will be fixed, and then the generator will be trained.
Correct answer: C
4. Reinforcement learning consists of environment, actions and rewards. The goal of reinforcement learning is to maximize the expectation of total rewards from a series of decisions made.
Correct answer: Correct
Final test questions (HCIA-AI V3.5 mock exam) (the questions and options are out of order):
Multiple choice questions:
1. The input image size is 200×200, and it goes through one layer of convolution (kernel size 5x5, padding 1, stride 2), pooling (kernel size 3x3, padding 0, stride 1), and another layer of convolution (kernel size 3x3 , padding 1, stride 1), which of the following options is the output feature map size?
A.95 B.96 C.97 D.98
Correct answer: C
2. Which of the following statements about RNN and LSTM is wrong?
A. If the input sequence is too long, the RNN model may suffer from the problem of gradient disappearance.
B.LSTM is a long short-term memory network, a time-cyclic neural network
C.LSTM is not suitable for processing important events with relatively long intervals and delays in time series
D. Based on LSTM, tasks such as language translation, robot control, document summarization, and chat robot control can be learned.
Correct answer: C
3. In the process of training a neural network, which value is continuously updated using the gradient descent method to minimize the loss function?
A.Hyperparameters
B. Eigenvalues
C.Number of samples
D.Parameters
Correct answer: D
4. The Relu function is often used in deep learning neural networks. Which of the following is the value range of this function?
A.[0,+∞)
B.[0,1]
C.[-1,1]
D.[-1,0]
Correct answer: A
5. Without considering regular terms, what points do the support vectors in SVM consist of?
A. Points falling on the dividing hyperplane
B. The point farthest from the dividing hyperplane
C. The closest point to the dividing hyperplane
D. Points of a certain category
Correct answer: C
6. The computer learns to distinguish which pictures are apples and which pictures are pears through labeled image data. Which of the following types of learning is most consistent with this scenario?
A. Supervised learning
B. Unsupervised learning
C. Semi-supervised learning
D. Reinforcement learning
Correct answer: A
7. Which of the following statements about classification models and regression models in machine learning is correct?
A. The prediction problem where the output variable is a finite number of discrete variables is a regression problem; the prediction problem where the output variable is a continuous variable is a classification problem.
B. For the evaluation of regression problems and classification problems, the most commonly used indicators are accuracy and recall.
C. Overfitting may occur in both regression problems and classification problems.
D. Logistic regression is a typical regression model.
Correct answer: C
8. The kernel function allows the algorithm to fit the largest hyperplane in the transformed high-dimensional feature space. Which of the following options is not a common kernel function?
A. Linear kernel function
B.Polynomial kernel function
C. Gaussian kernel function
D. Poisson kernel function
Correct answer: D
9. "ModelArts is a one-stop A development platform released by Huawei that only supports automatic learning and automatic annotation." Regarding the above description, which of the following statements is correct?
A. The description is correct, automatic learning and automatic annotation are the most important functions.
B. The description is correct. Automatic learning and automatic labeling can greatly improve the efficiency of AI projects.
C. This description is wrong. ModelArts supports automatic learning, preset models and code debugging.
D. The description is wrong. ModelArts only supports automatic learning, automatic annotation and code debugging.
Correct answer: C
10. Huawei Cloud Traffic Intelligence TrafficGo is a comprehensive urban traffic management solution. Which function is not implemented by traffic intelligence?
A. Traffic parameter perception
B. Multi-source data fusion
C. Real-time traffic detection
D. Assisted autonomous driving
Correct answer: D
11. Which of the following is not covered by Huawei Cloud Medical Intelligence?
A.Genome
B. Diagnosis and treatment
C.Clinical research
D.Drug research and development
Correct answer: B
12. Which of the following options is the interface used to save model weights in MindSpore?
A. mindspore.save_checkpoint(model, "model.ckpt")
B. mindspore.load_checkpoint("model.ckpt")
C. mindspore.load_param_into_net(model, param_dict)
D. mindspore.ops.no_gradient
Correct answer: A
13. During MindSpore training, in order to conveniently view the network model loss value, current training round time and other information
Which of the following options is required?
A.model.train
B.nn.ops
C.callback.LossMonitor
D.dataset.Mnist
Correct answer: C
14. Which of the following is not an artificial intelligence deep learning framework?
A.MindSpore
B.Scikit-learn
C.Pytorch
D.Caffe
Correct answer: B
15. What is used in TensorFlow to describe the calculation process?
A.Parameters
B. Conversation
C.Data flow diagram
D.Tensor
Correct answer: C
16. A manufacturer wants to produce a batch of virtual assistants for use in hospitals. For virtual assistants, voiceprint recognition mainly involves which of the following technologies?
A. Speech recognition and processing technology
B. Image recognition and processing technology
C. Expert system and knowledge graph technology
D.Image generation and enhancement technology
Correct answer: A
17. "Batch inference is a batch job that performs inference on batch data. There is no need to train the model before using batch inference." Regarding the above description, which of the following statements is correct?
A. This description is correct. Batch inference does not require further training.
B. The description is correct and inference means the training is over.
C. The description is wrong. The model must be trained before inference.
D. This description is wrong, batch inference does not require training operations.
Correct answer: C
18. Which of the following is not a cutting-edge application technology of artificial intelligence?
A. Diffusion model
B. Reinforcement learning
C.SDN
D.Unmanned driving
Correct answer: C
Multiple choice questions:
19. In deep learning, which of the following methods can reduce model overfitting?
A. Add more training samples
B. Dropout
C. Increase model complexity and improve the effect on the training set
D. Increase parameter penalty
Correct answer: ABD
20. Which of the following activation functions are prone to the vanishing gradient problem?
A.Sigmoid
B.Tanh
C.ReLU
D.Softplus
Correct answer: AB
21. The Adam optimizer can be regarded as a combination of which of the following?
A.Momentum
B. Adagrad
C.RMSprop
D.Nesterov
Correct answer:AC
22. Which of the following are hyperparameters that will be defined when training the model?
A. Training rounds (epoch)
B. Batch size
C. Learning rate
D.Loss
Correct answer: ABC
23. Which of the following options do the optimizers commonly used in machine learning include?
A. AdaGrad
B.SGD
C.ONNX
D.Adam
Correct answer: ABD
24. Which of the following options are included in the commonly used kernel functions in SVM?
A. Gaussian kernel function
B.Sigmiod kernel function
C. Linear kernel function
D.Polynomial kernel function
Correct answer: ABCD
25. Which of the following descriptions of data preprocessing are correct?
A. Data cleaning includes filling in missing values, finding and eliminating noisy data and outliers.
B. Data dimensionality reduction simplifies data attributes and avoids dimension explosion.
C. Data standardization reduces noise and improves model accuracy by standardizing data.
D. The final output of machine learning requires the help of a model, so training the model is more important than data preprocessing.
Correct answer: ABC
26. Which of the following options belong to the characteristics of the AI chip used on the terminal device?
A. High power consumption
B. High energy efficiency
C. Low latency
D.low cost
Correct answer: BCD
27. MindSpore’s key technologies that support device-edge-cloud on-demand collaboration include?
A. Unified model IR
B. Hard-software collaborative graph optimization technology
C. Device-cloud collaboration Federal Meta Learning
D. Centralized architecture
Correct answer: ABC
28. Which of the following options does MindSpore have for building Tensor?
A. Directly generated based on data
B. Generate from NumPy array
C. Generate directly from Scipy
D. Construct tensor using init initializer
Correct answer: ABD
29. Which of the following are common operations of Tensor in MindSpore?
A.switch()
B.size()
C.asnumpy(
D.max ()
Correct answer: BCD
30. Which of the following options are features of TensorFlow 2.x Eager Execution?
A.High performance
B.Deployability
C.Intuitive
D.Flexible
Correct answer: CD
31. Which of the following frameworks natively support distributed deep learning frameworks?
A.TensorFlow
B.MindSpore
C. CNDK
D.MXNet
Correct answer: ABD
32. Regarding the three major doctrines of artificial intelligence, which of the following statements are correct?
A. Symbolism favors theoretical logical reasoning; connectionism is weak in logic and explanation, and focuses on results.
B. Connectionism originates from bionics, especially the study of human brain models.
C. Behaviorism tends to apply practice and continuously learn from the environment to continuously correct actions.
D. Symbolism tends to apply practice and continuously learn from the environment to continuously correct actions.
Correct answer: ABC
33. Which of the current descriptions of AI application technology directions are correct?
A. Computer vision is the science of how to make computers "see".
B. Speech processing is a general term for various processing technologies that study the process of speech production, statistical characteristics of speech signals, speech recognition, machine synthesis, and speech perception.
C. Natural language processing is a discipline that uses computer technology to understand and use natural language.
D. Autonomous driving does not require the use of speech processing and computer vision technologies.
Correct answer: ABC
34. Only a more complete basic data service industry is the cornerstone for better development of AI technology. Which of the following options are correct regarding the data?
A. More secure data sharing is one of the cornerstones for the better development of AI technology.
B. Eliminating data barriers is important for the development of AI technology.
C. Eliminating the phenomenon of data islands is very important for the expansion of AI technology.
D. Data quality is not that important to the development of Al. Ensuring a large amount of data is enough.
Correct answer: ABC
True or False:
35. The loss function is used to evaluate the degree to which the predicted value of the model is different from the true value.
Correct answer: Correct
36. When training CNN, the input can be preprocessed such as rotation, translation, and scaling to improve the model's generalization ability.
Correct answer: Correct
37. The pooling layer can very effectively reduce the size of the parameter matrix, thereby reducing the number of parameters in subsequent convolutional layers.
Correct answer: False
38. Sigmoid is used in LSTM to implement threshold control, and TanH is used to implement data processing.
Correct answer: Correct
39. In the process of training a deep neural network with the Softsign activation function, if the network is very deep, more and more back-propagated gradients will fall into the saturation zone, making the gradient module smaller and smaller, eventually approaching 0. As a result, the weight cannot be updated.
Correct answer: Correct
40. The tanh function can solve the problem of vanishing gradient very well.
Correct answer: False
41. When the solution to the problem is complex, or the problem may involve a large amount of data but does not have a clear data distribution function, machine learning methods are more suitable.
Correct answer: Correct
42. When the hyperparameter search space is large, random search will be better than grid search.
Correct answer: Correct
43. Random search implements a random search for parameters, where each setting samples from the distribution of possible parameter values in an attempt to find the best subset of parameters.
Correct answer: Correct
44. Suppose we have a data set that gives the living area and price of 21,613 houses from house sales in a certain city. We can use a classification model to predict unknown prices.
Correct answer: False
45. Huawei Cloud EI provides industry intelligence, AI development platforms, and general AI capabilities.
Correct answer: Correct
46. MindSpore’s vision and value is to lower the threshold for AI development in the industry, release the computing power of Ascend chips, and help promote inclusive AI.
Correct answer: Correct
47. HMS Core currently supports HarmonyOS and Android, but does not support Windows.
Correct answer: False
48. Tensor is the basic data structure in MindSpore network operations. When MindSpore is running, all data will be converted into Tensor for calculation.
Correct answer: Correct
49. MindSpore supports distributed training and supports single-machine multi-card and multi-machine multi-card mode operation.
Correct answer: Correct
50. In MindSpore, mindspore.nn.Linear can be used to create a fully connected layer network.
Correct answer: False
51. When creating a network through the following code, you can add the decorator ms_function in the _init_ method to convert the model into static graph mode to obtain better performance.
Correct answer: False
52. Smart speakers are one of the typical application scenarios of voice processing.
Correct answer: Correct
53. On the premise of ensuring data privacy and security, federated learning uses different data sources to cooperatively train models to further break through data bottlenecks.
Correct answer: Correct
54. In a neural network based on connectionism, each node can express a specific meaning.
Correct answer: False
Fill in the blanks:
55. In a certain convolution layer of the convolutional neural network, assuming that there are 128 3*3 convolution kernels and the input feature map size is 28*28*64, the depth of the convolution kernel is ______. (please fill in Arabic numerals)
Correct answer: 64
56. In neural networks, the full Chinese name of CNN is ________.
Correct answer: convolutional neural network
57. The K value in KNN is artificially set and is a _______ of the model (input Chinese)
Correct answer: hyperparameters
58. Among the sub-services of Huawei Cloud Natural Language Processing Service, the ______ service provides users with fast and accurate translation services, helps users communicate across languages, and can be used in scenarios such as document translation. (please fill in Chinese)
Correct answer: machine translation
59. In the operation processing flow of MindSpore, int numbers in Python will be converted to the defined int64 type, and float numbers will be converted to the defined ______ type. (please fill in Chinese)
Correct answer: float32
60. The four elements of artificial intelligence are data, computing power, ___ (please fill in Chinese) and scenarios.
Correct answer: Algorithm