in Huawei

Artificial Intelligence Hauwei Questions

2k Views 1 Comment

  1. The Tensor Boost Engine (TBE) is a processor enablement layer built by Huawei for deep neural networks and Ascend AI processor operators library and a highly automated operators development kits.

True

False

  1. In Tensorflow 2.0 tf.data contains method to write data in TFRecord Format .

True

False

  1. Which of the following statement are through about hidden layers
  2. The more hidden layer the stronger the identification capability
  3. Excessive hidden layer may cause overfitting
  4. The more hidden layer the better
  5. Insufficient hidden layer may cause overfitting

Ans: AB

 

  1. which of the following are common kernel functions in support of vector machine (SVM)

A.Gausian kernel function

B.Sigmoid kernel function

C.Relu kernel function

D.Polynomial kernel function

Ans : ABD

 

5 which of the following development modes are supported by the ModelArts training platform

  1. In- Cloud development (codelab)
  2. In- Cloud development (Notebook +SDK)
  3. On- premises development (PyCharm +PyCharm toolkit)
  4. On – premises development (IDE+SDK)

Ans: ABCD

 

  1. Backed by HUAWEI CLOUD ' accumulated knowledge and expertise in AI Big data and other cutting edge technologies, GeoGenius offers a one stop AI development cloud platform for remote sensing data processing mining and management. Which of the following are the main functions of GeoGenus?
  2. Wheather forecasting
  3. Afforestation
  4. Emergency response and disaster prevention
  5. Smart transport infrastructure

Ans: AC

 

  1. Sensors and IOT technologies support for the () of the four element of AI data;Data;DATA

 

  1. The average value of tensors generated using tf.random.normal ([4], 2,1, tf.float32) is (). (Fill the blank with a number with a number). 2

 

  1. In mindspore if you want to define a function for collecting information after each step is complete, you need to inherit the () class. Callback; callback
  2. HUAWEI Cloud () helps enterprises build invoice management systems and self service reimburstment systems. ( Fill in the blank with the abbreviation of the service name). OCR; ocr

 

  1. In an ADAGRAD optimiser, the initial value of the cummulative square gradient is () (Fill in the blank with a number). 0

 

  1. MindSpore.ops.Gradoperation is the first order derivative method in Mindspore to compute gradients. It contains multiple parameters. When get_all is set to (), the derivatives of all parameters are computed. True

 

  1. The spartial relationship of image is that the local pixels are closely related and the pixels that are far away from each other are weakly related.

True

false

 

  1. The tree model used at the bottom layer of the random forest and GBDT algorithm is only the classification and regression Tree (CART).

True

false.

 

  1. Typically, a larger K value in the K nearest neighbour (KNN) algorithm reduces the impact of noise on classification but make the boundary between classes less obvious.

True

false.

 

  1. if a 32×32 image (without padding) is input and 5 x 5 kernel is used for convolution with a step of 2, the size of the output image is 13 x 13.

True

false

 

  1. The model Arts inference platform implements fast and efficient model inference and image recognition.

True

False    

 

  1. Grid search randomly selects hyper parameter combinations therefore, it has high efficient and accuracy.

True

False  

 

  1. The softplus function has a continous derivative and defines a smooth curved surface.

True

False 

 

  1. A decision selects a label from features of the training data to work as the node splitting standard. Different label selection standards generate different decision tree algorithm.

True

False

 

  1. Assuming the number of hyperparameter is the same, stochastic descent (SGD) combined with manual adjustment achieves a better effect than aqdaptive learning rates.

True

False

 

22.Filter methods for feature selection are restricted they do not tend to select redundant variables  because they consider the relationship between features.

True

False

 

  1. A model constructed using machine learning algorithms can be a functions that conforms to actual data distribution.

True

False

 

  1. Creative work, such as writing, will not be easily replicated by AI

True

False

 

  1. In the deep neural network, the back propagation algorithm uses the L&3×27; rule to perform layer by layer derivation and find the optimal parameters

True

False

 

  1. Device, edge cloud can respectively refers to smartphone . Huawei intelligent edge devices and Huawei CLOUD

True

False

 

  1. The XOR problem can be solved by using the deep neural network.

True

False

 

  1. Mindspore accelerated model convergence through automatic parallelism and second order optimization

True

False

 

  1. The PIL module in Tensorflow 2.0 is used to perform operation on images

True

False

 

  1. HUAWEI 3.0 supports the development of P50 smartphones

True

False

 

  1. Powered by Ascend AI processors, the Huawei Atlas AI computing platforms delivers product that come in various forms, such as module, card, edge statioons, secure and clusters, to build AI solutions for all scenarios across the device edge and processor

True

False

 

  1. Which of the followings statement is true about the low latency design of CPU
  2. The low clock frequency reduces the latency
  3. Complex logic control units increase the latency

C . A CPU is equipped with few ALU’s and caches. These cache merge access request to the same data and then access the DRAM reducing the latency

D.The powerful ALU can complete the computation in a short clock cycle.

Ans: D

 

  1. Which of the following statement is false about the ReLU function.
  2. The ReLU function is not differentiable at X=0 and a derivative is forcibly defined at this point.
  3. The ReLU function effectively alleviates the vanishing gradient problem
  4. The surface defined at the zero point of the ReLU function is not smooth enough in some regression problems
  5. Compared with sigmoid and tahn, the convergence of the ReLU function is slow

Ans: D

 

  1. Which of the following statement is false about gradient descent algorithm.
  2. The global gradient descent is more stable than the stochastic gradient descent (SGD)
  3. Whwn there are too many samples and GPU are not used for parallel computing , the convergence process of the global descent is time consuming.
  4. When GPU’S are not used for parallel computing, the Mini- Batch Gradient Decsent (MGBD take less time than SGD to complete an epoch
  5. Each time the global descent updates its weight, all training samples ne to be calculated.

Ans: C

 

  1. " knowledge representation is the unique method of representing knowlegde using a set of symbols in a structure that can be understood by compute squot; which of the following is true about this statement.
  2. This statement is false: The method of representing knowledge is not unique
  3. This statement is correct: The coding method of knowledge representation is unique
  4. This statement is correct: The knowledge representation can support expert system
  5. This statement is false: Knowledge representation cannot be used for expert rules or fuzzy inference

Ans: A

 

  1. Which of the followings is the correct shape of sensor ([[0,1) (2,3]] (4,5] (6, 77)]]?

A.(3, 3, 2)

  1. (3,2,4)
  2. (2,3,4)
  3. (2,2,2)

Ans: D

 

  1. Which of the following statements is true about variance & bias
  2. A model with low bias but high variance has high robustness
  3. A model with high bias but low variance has high precision
  4. A model with low bias and variance has low precision
  5. A model with high bias and variance performs poorly and will not be used.

Ans: D

 

  1. Which of the following statement is true about the ReLU function.
  2. The ReLU function is differentiable at X = 0 and the derivatives is 1
  3. Neuron nerosis does not
  4. There is no upper bound, so training is relatively easy to diverge
  5. The gradient is always 1 and the vanishing gradient problem can be perfectly solved

Ans: C

 

  1. Which of the following is false about the softmax function?
  2. The softmax function turns a vector of K real number into another vector of K real number
  3. The value of each element in the vector obtained from the softmax function ranges from 0 to 1
  4. The elements of the vector obtained from the softmax function do not add up to 1
  5. The softmax function is often used as the output layer of a multiclass classification task

Ans: C

 

  1. A generative adversarial network is like a game system. The generator produces fake samples, and the discriminator tries to distinguish red data from the data created by the generator. What is the ideal result.
  2. The generator is trained to fool the discrimitor
  3. The discriminator cannot distinguish real data from the data created by the generator
  4. The discriminator distinguishes real data from the data created by the generator
  5. Develop a high precision discriminator and a generator that cannot fool the discriminator

Ans: B

 

  1. Which of the following is the main function of HUAWEI CLOUD GeoGenius?
  2. Ecological detectin
  3. Heat demand forecast
  4. Real- time traffic detection
  5. Traffic Incident detection

 

  1. Which of the followings is the reasoning method of the production systems that draws conclusion through a rule library.
  2. Forward
  3. Backward
  4. Bidirectional
  5. Random

Ans: A

 

  1. Dirty data refers to data with quality problems. Which of the following statements is false about the data quality
  2. Noise: Data contains incorrect record or exceptions.
  3. Incompleteness: Data lacks attributes or contains some missing values
  4. Unprocessd: Data for which feature engineering has not been performed
  5. Inconsisitency: Data contains conflicting records

Ans: C

 

  1. Which of the followings is not a main function of Huawei CLOUD GeoGenius?
  2. Natural resource survey
  3. Ecological environment detection
  4. Weather forecast
  5. Real-Time traffic detections

Ans: D

 

  1. Which of the following statement is true about the loss functions typically used in deep learning
  2. Quadratic cost functions are usually for classification while cross entropy cost functions are used for clustering.
  3. The quadratic cost function reflects the gap between the target output and the actual outout
  4. The quadratic cost function reflects the gap between two probability distributions
  5. The purpose of training is to minimize the loss function.

Ans: B

 

  1. Which of the following is not a network operator in Mindspore
  2. Conv2D
  3. SGD
  4. Control Depend
  5. Softmax

Ans: C

 

  1. Which of the following product can be used by a company that wants to provide intelligent customer service?
  2. RDB
  3. Conversational BOT
  4. OBS
  5. GES

Ans: B

 

  1. Which of the following statement is true about the chip enablement layer in the Acsend AI software?
  2. The Ascend AI software stack encapsulates the framework involking and offline model generation capabilities .
  3. The Ascend AI software stack is the hardware compute basis of Ascend AI processors
  4. The Ascend AI software stack is nearest to the underlying compute resources and output operator layer task and the hardware.
  5. The Ascend AI software stack can invoke on appropriate deep learning framework based on the factors of related algorithms.

Ans: C

 

  1. Which of the following statements is true about rule based approaches?
  2. Rules are obtained through machine learning
  3. Rules based approaches use explicit programimg to solve problems
  4. Rule based approaches use sample for training
  5. Decision making rules are complex or difficult to desribe

Ans: B

 

  1. Which of the following statements are true about ensemble learning?
  2. Ensemble learning uses the decision tree as the bottom layer algorithm
  3. The voting policy of ensemble learning can be the averaging method
  4. A batch of features are randomly selected for subtree training in a random forest
  5. Xgboost supports parallel tarining like the random forest.

Ans: BC

 

  1. Which of the following statements are false about the Gated Recurrent Unit (GRU)
  2. Unlike long short term memory, GRU merges the cell state and hidden state.
  3. GRU is a variant of convolutionary Neural network (CNN)
  4. GRU combines the forget gate and update gates into a single input gate.
  5. GRU assembles the output of the previoud time step with the input of the current time step.

Ans: BC

 

  1. Which of the following use the Da Vinci architecture?
  2. Ascend 310
  3. RTX 3080
  4. Kunpeng 920
  5. Ascend 910

Ans: AD

 

  1. Which of the following statement are false about the universal engine of the Ascend 310 software stack.
  2. Provides general capabilities needed for neural network training.
  3. Provides compute resources and executes specific computing tasks
  4. Provides compute resources for Ascend AI
  5. Outputs tasks at the operation layer for hardware

Ans: ABCD

 

  1. Which of the following are part o minspore
  2. Mind Expression API layer
  3. Mind compile’s real-time graph – level compilation subsystem .
  4. CUDA and third party chips
  5. MindData

Ans: ABD

 

  1. Which of the followings statement are false about common optimizers
  2. One of the advantage of Adagrad optimizers is that the parameters update operation does not end too early
  3. When the momentum optimizer is used, parameter are updated by using the same learning rate, but momentum coefficient keep changing with each iteration
  4. Adam sets different learning rates for different parameter
  5. The RMSPROP Optimizer introduces an attenuation coefficient to ensure a certain proportion of gradient attenuation in each round.

Ans: AB

 

  1. Which of the following are steps of the BPTT algorithm
  2. Calculating the output of each neuron (forward)
  3. Calculating the error of each neuron (backward)
  4. Calculating the gradient of each weight
  5. Calculating the layer of neuron

Ans: ABCD

 

  1. Which of the following statement are true about the mindspore componets
  2. Communication: Processes data flow communication between the CPU and memory
  3. Dataset: Dataset processing module that loads and preprocesses data

C: NN: Neural Network cells in Mindspore that define loss functions and optimizes

D: Train: relates to training model and model quantization module

Ans: BC

 

  1. In Tensor flow2.0 which of the following cannot be used to create all zero tensor?
  2. tf.zeros ()
  3. tf.fill ( (1,3,0)
  4. tf.zeros –frame()
  5. tf.zeros tuple()

Ans: CD

 

  1. The accuracy of a trained model is high on a training data set but low on a test dataset. In this case which of the followings operations can be perfomed
  2. Use the sigmoid activation function
  3. Perform data asymetric
  4. Add regular terms

Ans: BC

 

  1. Which of the followings are major advantage of kera
  2. Easy to use
  3. Low level coding style
  4. Modular and combination
  5. Easy to

Ans: ACD

 

  1. Which of the following statements are true about common activation functions in deep learning.
  2. The tahn function is symetric with respect to the origin and the mean of its ouput is closer to 0
  3. During training of a deep neural network, the sigmoid, tanh and softsign functions cannot prevent vanishing gradient

Ans: AB

 

What do you think?

1 Point
Upvote Downvote

Written by admin

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *

GIPHY App Key not set. Please check settings

One Comment

Loading…

Artificial Intelligence Mock Study Questions

HUAWEI – Questions & Answer 2