神經網絡期末復習資料（題庫）

鏈接：https://blog.csdn.net/Pqf18064375973/article/details/148332887?sharetype=blogdetail&sharerId=148332887&sharerefer=PC&sharesource=Pqf18064375973&sharefrom=mp_from_link

【測試】

• The —— does not only work accordina to the aldorithm but also can predict a solution for a #&task and make conclusions using its previous experience
  A.Artificial Intelligence
  B.Neural Network
  C.Deep Neural Network
  D.Machine learning
• Deep learning is a subset of machine learning which is essentially a —— with 3 or more layers.
  A.Machine Learning
  B.Artificial intelligence
  C.Neural network
  D.Big data
• The —— is this kind of technology that is not an algorithm. it is a network that has weights on it. and you can adiust the weights so thatit leamns. You teach i
  through trials.
  A.Machine Learning
  B.Artificial intelligence
  C.Deep learning
  D.Neural network
• Team of Sir Geoffrey Hinton, also dubbed as “The Father of Deep Learning”, published the research paper on ——
  A.forward propagation
  B.back propagation
  C.deep propagation
  D.accurate propagation
• —— is a type of artificial neural network which uses sequential data or time series data.
  A.Restricted Boltzmann Machines (RBMs)
  B.Deep Belief Networks (DBNs)
  C.Convolutional Neural Networks (CNNs)
  D.Recurrent Neural Networks (RNNs)
• Which optimizer has a fixed learning rate?
  A.AdaDelta
  B.RMSprop
  C.Gradient Descent
  D.Adam
• Some approaches to machine learning tend to focus on learning only one or two layers of representations ofthe data, hence,theyre sometimes callec ——
  A.Shallow Learning
  B.Broad Learning
  C.Deep learning
  D.Neural network
• Which optimizer performs well on sparse data?
  A.Adam
  B.AdaGrad
  C.RMSprop
  D.Momentum
• This deep learning algorithm became very popular after the Netflix Competition whereratings for movies and beat most of its competition —— was used as a collaborative filtering technique to predict user
  A.Restricted Boltzmann Machines (RBMs)
  B.Deep Belief Networks (DBNs)
  C.Convolutional Neural Networks (CNNs)
  D.Recurrent Neural Networks(RNNS
• lmage caption is the example of which type of RNN?
  A.One to One RNN
  B.One to Many RNN
  C.Many to One RNN
  D.Many to Many RNN
• What is the output range of the ReLU activation function?
  A.(1,1)
  B.[0, ∞)
  C.(0, 1)
  D.(-∞,∞)
• Operations like data cleaning to find missing values ,to remove useless data and perform basic statistical analysis like drawing plots, comparing diferent features of the data set and more. This is done in which phase of life cycle
  A.Data Acquisition
  B.Data exploration
  C.Modelling
  D Fvaluation
• How well does the Transformer model handle long-term dependencies?
  A.general
  B.poor
  C.excellent
  D.none
• —— algorithms use historical data as input to predict new output values.
  A.Machine Learning
  B.Artificial intelligence
  C.Deep learning
  D.Neural network
• Which RNN variant can handle long-term dependencies and is faster?
  A.GRU
  B.LSTM
  C.SimpleRNN
  D.Bidirectional LSTM
• —— are algorithms or methods used to change the attributes of the neural network such as weights 換and learning rate to reduce the losses
  A.0ptimizers
  B.Neural network program
  C.Matrix
  D.None of these
• In visual system of mammalian,peripheral part is formed by the eves,theintermediate (by the transmission of nerve impulses -the optic nerves,and the central-the visual centers in the
  A.cerebral cortex.
  B.ears
  C.pulse
  D.shape
• Mammalian vision is the process of mammals that
  A.Perceiving light,
  B.Analyzing it
  C.Forming subjective sensations
  D.All of these
• The neural network is this kind of technolgy that is not an alorithm. it is a networkthat has weihts on it and you can adiust the weights so that it leams, You teach it through trials.This Definition is said by
  A.Sir Geoffrey Hinton
  B.Howard Rheingold
  C.Mike caferella
  D.Matei Zaharia
• CNN layers are
  A.Convolutional Layer
  B.Pooling Layer
  C.Fully connected layer
  D.All of these
• The term is frequently applied to the project of developing systems endowed with the intellectual processes characteristic ofhumans, such as the ability to reason，discover meaning, generalize, or learn from past experience.
  A.Machine Learning
  B.Artificial intelligence
  C.Deep learning
  D.Neural network
• Which optimizer combines momentum and adaptive learning rate?
  A.Momentum
  B.Adam
  C.AdaGrad
  D.SGD
• The——is the fundamental building block of neural networks
  A.Artificial Intelligence
  B.Neuron
  C.Deep learning
  D.Machine learning
• CNN has a.that has several filters to perform the convolution operation
  A.Convolution Layer
  B.Rectified Linear Unit (ReLU)
  C.Pooling Layer
  D.Fully Connected Layer
• What are the main advantages of the GELU activation function over ReLU?
  A.More efficient calculation
  B.Avoid gradient disappearance
  C.Larger output range
  D.Suitable for hidden layers
• During image processing first layer’s output is passed on to the next layer which detects more complex features such as ——
  edges
  A.combinational or sequentia!
  B.corners or diagonal
  C.combinational or corners
  D.vertical or corners
• A convolution tool that separates and identifies the various features of the image for analysis in a process called as——
  A.Feature identification
  B.Feature Extraction
  C.Partial Extraction
  D.Feature recognition
• —— show very effective results in image and video recognition, natural language processing, and recommender systems.
  A.Feed forward Neural Network- Artificial Neuron
  B.Radial Basis Function Neural Network
  C.Multilayer Perceptron
  D.Convolutional Neural Network
• —— is useful for regression, classification, dimensionality reduction, feature learning, topic modelling and collaborative filtering
  A.Restricted Boltzmann Machines (RBMs)
  B.Deep Belief Networks (DBNs)
  C.Convolutional Neural Networks (CNNs)
  D.Recurrent Neural Networks (RNNs)
• Which activation function has the “Dying ReLU” problem?
  A.Leaky ReLU
  B.ReLU
  C.Tanh
  D.Sigmoid
• In Visual system of mammalian, the capabilities of the nervous system to process informaton received are limited to
  A.tens of bits per second.
  B.hundreds of bits per second.
  C.thousands of bits per second
  D.tens of bytes per second.
• Modern deep learning often involvesof successive layers of representations
  A.1 or 100
  B.100 or 1000
  C.10 or 100
  D.1 or 2
• Which optimizer can speed up when dealing with plateau problems?
  A.SGD
  B.Momentum
  C.AdaGrad
  D.Gradient Descent
• Team of—— , also dubbed as “The Father of Deep Learning”, published the research paper on Back propagation
  A.Mike caferella
  B.Matei Zaharia
  C.Sir Geoffrey clinton
  D.Sir Geoffrey Hinton
• What scenario is the Softmax activation function typically used for?
  A.Return to mission
  B.Two-category output layer
  C.Multi-classification output layer
  D.Hidden layer
• Which Phase of deep learning life cycle , gather data from reliable data sources
  A.Problem scoping
  B.Data Acquisition
  C.Data exploration
  D.Modelling
• —— basically extends the area of an image in which a convolutional neural network processes
  A.Padding
  B.stride in CNN
  C.ANN
  D.Neural Network
• Which phase of deep learning life cycle takes care about spelling mistakes or maybe labelling the data wrong.
  A.Problem scoping
  B.Data Acquisition
  C.Data exploration
  D.Modelling
• Full form of deep learning model RBM is ——
  A.Real Boltzmann Machines
  B.Restricted Bussiness Machines
  C.Restricted Boltzmann Machines
  D.Restricted Boltzmann Major
• fraud detection, spam filtering, malware threat detection, business process automation
  (BPA)and predictive maintenance are examples of
  A.Machine Learning
  B.Artificial intelligence
  C.Deep learning
  D.Neural network
  E
• The output range of the**【 sigmoid 】**activation function is (0, 1)
• The hyperparameters of the ADAM optimizer include 【 learning rate 】,B1, B2, and ε.
• In the RNN variant, the Use Case Complexity of Bidirectional LSTM is the**【High】**
• Yann LeCun built the first convolutional neural network called **【 LeNet 】**in 1988
• The first layer of image processing usually extracts basic features such as horizontal or diagonal 【edges】
• AdaDelta optimizers usually do not require a global learning rate.【T F 】
• The output of the Tanh activation function is zero-mean.【T F】
• SimpleRNN is suitable for dealing with long-term dependencies.【T F】
• Rectified Linear Unit (RelU) : CNN’s have a ReLU layer to perform several flters to perform the convolution operation.【T F】
• Responsible for this process in mammals is the touch sensory svstem,the foundations of which were formed at an eary stace in the evolution of chordates.【T F】

---

【課后題】

• —— the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings.

  a. Artificial intelligence

  b. Machine learning

  c. Deep learning

  d. None of the above

• Which of the following is not a type of Neural network

a. Modular-To-Modular Models.

b. Radial Basis Function Neural Network

c. Multilayer Perceptron

d. Convolutional Neural Network

• Operations like data cleaning to find missing values , to remove useless data and perform basic statistical analysis like drawing plots, comparing different features of the data set and more. This is done in which phase of life cycle

  a. Data Acquisition

  b. Data exploration

  c. Modelling

  d. Evaluation

• It is useful for regression, classification, dimensionality reduction, feature learning, topic modelling and collaborative filtering.

  a. Restricted Boltzmann Machines (RBMs)

  b. Deep Belief Networks (DBNs)

  c. Convolutional Neural Networks (CNNs)

  d. Recurrent Neural Networks (RNNs)

• CNN has a —— that has several filters to perform the convolution operation.

  a. Convolution Layer

  b. Rectified Linear Unit (ReLU)

  c. Pooling Layer

  d. Fully Connected Layer

• —— can be used to build speech-recognition, image-recognition, and machine-translation software.

  a. Deep Belief Networks (DBNs)

  b. Convolutional Neural Networks (CNNs)

  c. Recurrent Neural Networks (RNNs)

  d. Multilayer Perceptrons (MLPs)

• Yann LeCun, director of Facebook’s AI Research Group, is the pioneer of ——

  a. Restricted Boltzmann Machines (RBMs)

  b. Deep Belief Networks (DBNs)

  c. Convolutional Neural Networks (CNNs)

  d. Recurrent Neural Networks (RNNs)

• A ____________ that utilizes the output from the convolution process and predicts the class of the image based on the features extracted in previous stages.

  a. Convolution Layer

  b. Rectified Linear Unit (ReLU)

  c. Pooling Layer

  d. Fully Connected Layer

• Mammalian vision is the process of mammals that

  a. Perceiving light,

  b. Analyzing it

  c. Forming subjective sensations

  d. All of these

• A deep CNN model consists of a finite set of processing layers that can learn various features of input data (e.g., image) with _____ level of abstraction.

  a. single

  b. multiple

  c. one

  d. two

• —— basically extends the area of an image in which a convolutional neural network processes.

  a. Padding

  b. stride in CNN

  c. ANN

  d. Neural Network

• Advantages of convolution layers are:

  a. Sparse Connectivity

  b. Weight Sharing

  c. A & B

  d. none of these

• Rosenblatt’s perceptron is basically a classifier.

  a. unary

  b. binary

  c. ternary

  d. All of these

• The ______ is the fundamental building block of neural networks.

  a. Artificial Intelligence

  b. Neuron

  c. Deep learning

  d. Machine learning

• It is one kind of backpropagation network which produces a mapping of a static input for static output.

  a. Static back-propagation

  b. Dynamic back-propagation

  c. Concurrent Backpropagation

  d. Recurrent Backpropagation

• The ______________ does not only work according to the algorithm but also can predict a solution for a task and make conclusions using its previous experience.

  a. Artificial Intelligence

  b. Neural Network

  c. Deep Neural Network

  d. Machine learning

• Apple’s Siri uses _________ for image recognition and voice recognition respectively.

  a. TensorFlow

  b. PyTorch

  c. Loss Function

  d. Deep Neural Network

• _______ can be used as an alternative to cross-entropy, which was initially developed to use with a support vector machine algorithm.

  a. Cross Entropy Loss

  b. Hinge Loss

  c. Squared Hinge Loss

  d. All of these

• ________ are algorithms or methods used to change the attributes of the neural network such as weights and learning rate to reduce the losses.

  a. Optimizers

  b. Neural network program

  c. Matrix

  d. None of these

• It is an extension of AdaGrad which tends to remove the decaying learning Rate problem of it.

  a. Mini-Batch Gradient Descent

  b. Momentum

  c. Adaptive Gradient Descent (AdaGrad)

  d. AdaDelta

• ____________ occurs when the derivative or slope will get smaller and smaller as we go backward with every layer during backpropagation.

  a. Adaptive Gradient Descent

  b. Vanishing Gradient

  c. Gradient Descent

  d. None of these

• Image caption is the example of which type of RNN?

  a. One to One RNN

  b. One to Many RNN

  c. Many to One RNN

  d. Many to Many RNN

• Full form of GRU is _______

  a. Gated Recurrent Unit

  b. Good Recurrent Universe

  c. Gated Recursive Unit

  d. None of these

• Recurrent neural networks lie in their diversity of application. When we are dealing with RNNs they have a great ability to deal with various input and output types like

  a. Sentiment Classification

  b. Image Captioning

  c. Language Translation

  d. All of these

• Which of the following is not an application of autoencoders?

  a. Image Coloring

  b. Feature variation

  c. Dimensionality Reduction

  d. Voice recognition

• A ______ is an unsupervised deep learning technique that helps a neural network encode unlabeled training data.

  a. Sparse autoencoders

  b. Contractive autoencoder

  c. Deep autoencoder

  d. All of these

• Which property of autoencoders describes the decompressed outputs will be degraded compared to the original inputs?

  a. Data Specific

  b. Lossy

  c. Lossless

  d. Algorithm specific