Hcia Ai Huawei Mock Exam Written Docu [PDF]

Citation preview

The speed of grid search must be faster than random search. Answer: False



The biggest difference between logic regression and perceptron is that logistic regression uses the Sigmoid activation function at the output layer, whereas perceptron uses the sign function. In addition, neither of the two models can resolve the XOR problem. Answer: True



HiLens can help small and micro companies realize smart access control. Answer: True



Polynomial regression cannot resolve the nonlinear problem. Answer: False



Softmax can’t be used for multiple classification problems. Answer: False



The Adam optimizer in TensorFlow 2.0 is the built-in optimizer of the Keras module. Answer: True



The HiAI mobile computing platform architecture supports various mainstream front-end frameworks, such as TensorFlow and Caffe. Answer: True



To install TensorFlow of the GPU version, you need to install CUDA and cuDNN. Answer: True



In a feedforward network, neurons in the same layer are connected to each other. Answer: False



Backpropagation makes gradient descent more efficient by means of dynamic programming. Answer: True



In a medical image recognition task, we usually are more concerned about the recall ratio than the precision ratio when evaluating model performance. Answer: True



After the image label service is called, only one corresponding label value is returned. Answer: False



Similar to hidden layer, the quantity of neurons at both an input layer and an output layer of a neural network is a hyperparameter that can be adjusted. Answer: False



Cifar10 is a built-in data set of Keras. Answer: True



The generalization error will continue to decrease as the complexity of the model increases. Answer: False



MindSpore supports GPU acceleration. Answer: True



Voice recognition refers to the recognition of text data as audio data. Answer: False



PyTorch supports GPU acceleration. Answer: True



In the KNN model, a small K value may cause model underfitting. Answer: False



Malformed data is not dirty data. Answer: False



Factory A produces 60% of an automotive company’s products, and factory B accounts for 40%. The defective rate of factory A is 1%, and that of factory B is 3%. If a vehicle is defective, what is the possibility that the vehicle is produced by factory A? Answer: B. 1/3



Which of the following statements about dropout is incorrect? Answer: D. dropout can’t be added during inference.



Which of the following reasons is the main reason why deep learning developed rapidly after 2010? Answer: D. Improvement of computing power



Which of the following methods cannot resolve overfitting in a decision tree? Answer: D. Use the integration algorithm to set up an AdaBoost model.



Which of the following network layers does not need to be included in a convolutional neural network? Answer: C. Pooling layer



Which of the following is not a feature of the Atlas 500 AI edge stations? Answer: C. High energy consumption



What determines the number of neurons in the input layer of the neural network? Answer: Data dimension



What operation does “isinstance” represent in TensorFlow 2.0? Answer: C. Determine whether it is Tensor



Which of the following schools does the expert system belong to? Answer: A. Symbolism



Input_shape = (4, 28, 28, 1) X = tf.random.normal(input_shape) Y = tf.keras.layers.Conv2D(filters=2, kernel_size=3, strides=(1, 1), padding=’valid’, activation=’relu’, Input_shape=input_shape[1:])(x) print(y.shape) What is the output of the preceding code? Answer: B. (4,26,26,2)



Without Padding, use a 3X3 convolution kernel to convolve a 5X5 image with stride 1. Which of the following options is the size of the output image? Answer: A. 3X3



When was the Turing Test proposed? Answer: A. 1950



Which of the following options is the architecture adopted by Huawei Atlas series chips? Answer: D. Da Vinci



In the process of training the neural network, which of the following methods is used to update the parameters so as to minimize the loss function? Answer: C. Gradient descent



Which of the following options is not the feature of ModelArts? Answer: C. Support automatic tuning



In machine learning, how many support vectors can the SVM algorithm have? Answer: D. Several



import numpy as np n = np.arange(24).reshape(2,-1,2,2) Which of the following is the return result of n.shape after running the preceding code? Answer: A. (2,3,2,2)



Which of the following options is the artificial intelligence development framework launched by Huawei? Answer: C. MindSpore



Which of the following options is a commonly used loss function for classification problems? Answer: B. Cross Entropy Loss Function



Which of the following problems can’t be handled by the perceptron? Answer: C. XOR



Which of the following models is not suitable for the scenario where the covariance between data features is large? Answer: C. Naive Bayes



Which of the following options is Huawei’s artificial intelligence computing platform? Answer: D. Atlas



A company wants to use the HiAI platform to develop its own APP. Which of the following IDEs can be used in conjunction with the HiAI platform? Answer: B. Android Studio



Which of the following options is the classification of the AI chip business from the application? Answer: C. Training and Inference



Which of the following options belong to the activation function of the neural network? Answer: A. Sigmoid B. Softmax C. Softplus D. ReLu



Which of the following statements about LSTM are correct? Answer: A. The LSTM unit has three types of gates: input gate, forget gate, and output gate. B. The input of one time sequence of LSTM is concatenated with the output of its last time sequence. C. If peephole is used in LSTM, the vector of the output and memory unit of the previous time sequence are concatenated with the input vector. D. In LSTM, if the dimension of the input vector is 200 and the dimension of the hidden layer is 100, and the peephole is not considered, the parameter quantity at this layer is calculated as follows: 4x ((100 + 200) X 100 + 100)



Which of the following options are the main advantages of the Keras interface in TensorFlow 2.0? Answer: A,B,C A. Convenient for users B. Modular and combinable C. Easy to expand



Which of the following options belong to the application scenarios of natural language processing? Answer: B & D B. Machine translation D. Public opinion monitoring



Which of the following may cause overfitting? Answer: B,C,D B. The model capacity is large C. The training set includes a large amount of dirty data D. The number of training sets is small



Which of the following are common Python computer vision libraries? Answer: C & D C. matplotlib D. cv2



Which of the following options belongs to Huawei HiAI platform content? Answer: A, B, D A. HiAI Service B. HiAI Foundation D. HiAI Engine



What are the elements of AI? Answer: A. Algorithms B. Computing power C. Data D. Scenarios



Which of the following are major functions of Mind Expression in MindSpore? Answer: A. Operator-level automatic differentiation B. Automatic parallelization C. Automatic operator generation D. Semi-automatic data labeling



Which of the following are the problems you may encounter when training a generative adversarial network? Answer: A. An overview learning rate leads to model divergence B. The discriminator is too well trained at the beginning of training. As a result, the generator cannot be trained. C. The mode collapse results in insufficiently diversified images. D. The number of data sets is small, causing model overfitting



Which of the following are the major modules of the Ascend AI processor? Answer: A. B. C. D.



Chip system control CPU AI computing engine Multi-level System-On-a-Chip cache or buffer Digital vision pre-processing module



Which of the following options are Huawei products? Answer: A. Atlas 200DK B. Ascend 310 C. Ascend 910



Which of the following are AI applications? Answer: A. B. C. D.



Smart speaker Smart customer service Autonomous driving Alpha Go



Which of the following statements about TensorFlow are correct? Answer: A. Keras is a high-level API of TensorFlow B. TensorFlow 1.0 uses static graphs C. TensorFlow 2.0 uses Eager Execution by default



Compared with the Sigmoid activation function, which of the following options are the advantages of ReLU? Answer: A. Mitigating the problem of vanishing gradient D. Save calculation



Which of the following are deep learning frameworks? Answer: A. B. C. D.



TensorFlow PyTorch MXNet MindSpore