Deep Learning and Reinforcement Learning Week 6 All Quiz

Practice: Recurrent Neural Networks 

1. (True/False) Recurrent Neural Networks are a class of neural networks that allow previous outputs to be used as inputs while having hidden states.

A) True
B) False

ANSWER= (A) True

 

2. (True/False) Recurrent Neural Networks are well suited in applications in which the context is important and needs to be incorporated in the prediction.

A) True
B) False

ANSWER= (A) True

 

3. These are the two main outputs of a recurrent neural network:

A) Prediction and state
B) Prediction and parameters
C) Prediction and recurrence
D) Prediction and learning rate

ANSWER= (A) Prediction and state

Practice: LSTM and GRU  

1. (True/False) The main motivation behind LSTM is to make it easier to keep information from distant past in current memory without reinforcement.

A) True
B) False

ANSWER= (A) True

 

2. RNNs are augmented with the following Gate Units:

A) Enter gate, leave gate, print gate
B) Input gate, forget gate, output gate
C) Store gate, remove gate, feed forward gate
D) Recursive gate, keep gate, reinforce gate

ANSWER= (B) Input gate, forget gate, output gate

 

3. Select the correct assertion regarding the gate units of RNNs:

A) The gate units control how long the events will stay in memory.
B) The gate units control if the events will stay in memory.
C) The gate units control how many items can be stored in memory.
D) A and B

ANSWER= (D) A and B

 

Practice: Regularization  

1. Which regularization technique can shrink the coefficients of the less important features to zero?

A) L2
B) Dropout
C) Batch Normalization
D) L1

ANSWER= (D) L1

 

2. (True/False) Batch Normalization tackles the internal covariate shift issue by always normalizing the input signals, thus accelerating the training of deep neural nets and increasing the generalization power of the networks.

A) True
B) False

ANSWER= (A) True

 

3. Regularization is used to mitigate which issue in model training?

A) Underfitting
B) High bias and low variance
C) Overfitting
D) Both underfitting and overfitting

ANSWER= (C) Overfitting

 

Week 6 Final Quiz  

1. (True/False) RNN models are mostly used in the fields of natural language processing and speech recognition.

A) True
B) False

ANSWER= (A) True

 

2. (True/False) GRUs and LSTM are a way to deal with the vanishing gradient problem encountered by RNNs.

A) True
B) False

ANSWER= (A) True

 

3. (True/False) GRUs will generally perform about as well as LSTMs with shorter training time, especially for smaller datasets.

A) True
B) False

ANSWER= (A) True

 

4. (True/False) The main idea of Seq2Seq models is to improve accuracy by keeping necessary information in the hidden state from one sequence to the next.

A) True
B) False

ANSWER= (A) True

 

5. (True/False) The main parts of a Seq2Seq model are: an encoder, a hidden state, a sequence state, and a decoder.

A) True
B) False

ANSWER= (B) False

 

6. Select the correct option, in the context of Seq2Seq models:

A) The Greedy Search algorithm selects one best candidate as an input sequence for each time step while the Beam Search produces multiple different hypothesis based on the output from the encoder.
B) The Beam Search algorithm selects one best candidate as an input sequence for each time step while the Greedy Search produces multiple different hypothesis based onthe output from the encoder.
C) The Greedy Search algorithm selects one best candidate as an input sequence for each time step while the Beam Search produces multiple different hypothesis based on conditional probability.
D) The Beam Search algorithm selects one best candidate as an input sequence for each time step while the Greedy Search produces multiple different hypothesis based on conditional probability.

ANSWER= (C) The Greedy Search algorithm selects one best candidate as an input sequence for each time step while the Beam Search produces multiple different hypothesis based on conditional probability.

 

7. Which is the gating mechanism for RNNs that include a reset gate and an update gate?

A) GRUs
B) LSTMs
C) Refined Gate
D) Complex Gate

ANSWER= (A) GRUs

8. LSTM models are among the most common Deep Learning models used in forecasting. These are other common uses of LSTM models, except:

A) Speech Recognition
B) Machine Translation
C) Image Captioning
D) Generating Images
E) Anomaly Detection
F) Robotic Control

ANSWER= (D) Generating Images

Deep Learning and Reinforcement Learning Week 5 All Quiz

Practice: Transfer Learning 

1. The main idea of transfer learning of a neural network is:

A) To keep the early layers of a pre-trained network and re-train the later layers for a specific application.
B) To use the early layers to capture features that are more particular to the specific data you are trying to classify.
C) To train the early layers such that their weights have a higher impact on the final result.
D) To re-train the early layers for a specific application and transfer it to a different data set

ANSWER= (A) To keep the early layers of a pre-trained network and re-train the later layers for a specific application.

 

2. In the context of transfer learning, which is a guiding principle of fine tuning?

A) Fine tuning the hyperparameters of the CNNs
B) Using data that is similar to the pre-trained network
C) Adjust the weights of the neural network
D) Increase the number of later layers iteratively

ANSWER= (B) Using data that is similar to the pre-trained network

 

3. In the context of transfer learning, what do we call the process in which you only train the last or a few layers instead of all layers of a neural network?

A) Frozen layers
B) Frozen weights
C) Updated learning
D) Updated layers

ANSWER= (A) Frozen layers

Practice: Convolutional Neural Network Architectures  

1. This concept came as a solution to CNNs in which each layer is turned into branches of convolutions:

A) Inception
B) Workload portion
C) Hebbian Principle
D) Network Concatenation

ANSWER= (A) Inception

 

2. Which CNN Architecture is considered the flash point for modern Deep Learning?

A) AlexNet
B) VGG
C) Inception
D) ResNet
E) LeNet

ANSWER= (A) AlexNet

 

3. Which CNN Architecture can be described as a "simplified, deeper LeNet" in which the more layers, the better?

A) Deep Lenet
B) AlexNet
C) VGG
D) Inception
E) ResNet

ANSWER= (C) VGG

4. Which CNN Architecture is the precursor of using convolutions to obtain better features and was first used to solve the MNIST data set?

A) AlexNet
B) VGG
C) Inception
D) ResNet
E) LeNet

ANSWER= (E) LeNet

 

5. The motivation behind this CNN Architecture was to solve the inability of deep neural networks to fit or overfit the training data better when adding layers.

A) LeNet
B) AlexNet
C) VGG
D) Inception
E) ResNet

ANSWER= (E) ResNet

 

6. This CNN Architecture keeps passing both the initial unchanged information and the transformed information to the next layer.

A) LeNet
B) AlexNet
C) VGG
D) Inception
E) ResNet

ANSWER= (E) ResNet

 

7. Which activation function was notably used in AlexNet and contributed to its success?

A) ReLU (Rectified Linear Unit)
B) Sigmoid
C) Tanh
D) Leaky ReLU

ANSWER= (A) ReLU (Rectified Linear Unit)

 

Practice: Regularization  

1. Which regularization technique can shrink the coefficients of the less important features to zero?

A) L2
B) Dropout
C) Batch Normalization
D) L1

ANSWER= (D) L1

 

2. (True/False) Batch Normalization tackles the internal covariate shift issue by always normalizing the input signals, thus accelerating the training of deep neural nets and increasing the generalization power of the networks.

A) True
B) False

ANSWER= (A) True

 

3. Regularization is used to mitigate which issue in model training?

A) Underfitting
B) High bias and low variance
C) Overfitting
D) Both underfitting and overfitting

ANSWER= (C) Overfitting

 

Week 5 Final Quiz  

1. (True/False) In Keras, the Dropout layer has an argument called rate, which is a probability that represents how often we want to invoke the layer in the training.

A) True
B) False

ANSWER= (B) False

 

2. What is a benefit of applying transfer learning to neural networks?

A) Train early layers for specific applications and generalize that with later pre-trained layers.
B) Save early layers for generalization before re-training later layers for specific applications.
C) Easily adjust weights of early layers to reduce training time.
D) Place heavy focus on training layers that generalize the model.

ANSWER= (B) Save early layers for generalization before re-training later layers for specific applications.

 

3. By setting ` layer.trainable = False` for certain layers in a neural network, we____

A) set the layers’ weights to zero
B) exclude the layers during training because they should be discarded
C) freeze the layers such thattheir weights change synchronously during training.
D) freeze the layers such that their weights don’t update during training.

ANSWER= (D) freeze the layers such that their weights don’t update during training.

 

4.Which option correctly orders the steps of implementing transfer learning?
1. Freeze the early layers of the pre-trained model.
2. Improve the model by fine-tuning.
3. Train the model with a new output layer in place.
4. Select a pre-trained model as the base of our training.

A) 3, 1, 2, 4
B) 3, 2, 4, 1
A) 4, 2, 3, 1
B) 4, 1, 3, 2

ANSWER= (B) 4, 1, 3, 2

 

5. Given a 100x100 pixels RGB image, there are _____ features.

A) 300
B) 100
C) 30000
D) 10000

ANSWER= (C) 30000

 

6. Before a CNN is ready for classifying images, what layer must we add as the last?

A) Dense layer with the number of units corresponding to the number of classes
B) Flattening layer with the number of units corresponding to the number of classes
C) Flattening layer with the number of units corresponding to (number of classes*input size)
D) Dense layer with the number of units corresponding to (number of classes*input size)

ANSWER= (A) Dense layer with the number of units corresponding to the number of classes

 

7. In a CNN, the depth of a layer corresponds to the number of:

A) filters applied
B) input layers
C) channel-filter combinations
D) color channels

ANSWER= (A) filters applied

Deep Learning and Reinforcement Learning Week 4 All Quiz

Practice: Convolutional Neural Networks 

1. Given the syntax below, select the option that will best improve a CNN model that you are trying to fit.
1. model.fit(x_train, y_train, batch_size=batch_size, epochs=100, validation_data=(x_test, y_test))

A) Remove the validation_data option.
B) Increase the number of epochs to 100.
C) ecrease the number of epochs to 50.
D) Add shuffling, by adding “, shuffle=True” at the end.

ANSWER= (D) Add shuffling, by adding “, shuffle=True” at the end.

 

2. Which of the following statements is TRUE about a kernel in a Convolutional Layer applied to an image?

A) Kernels allow the convolutional layers to perform nonlinear transformations.
B) Kernels detect local features in an image such as lines, corners, and edges.
C) Kernels identify which channel in the input data contains the most information.
D) Kernels ease computation by reducing the number of dimensions in an image that must be processed.

ANSWER= (B) Kernels detect local features in an image such as lines, corners, and edges.

Week 4 Final Quiz  

1. What is the main function of backpropagation when training a Neural Network?

A) Preprocess the input layer
B) Make adjustments to the weights
C) Make adjustments to the loss function
D) Propagate the output on the output layer

ANSWER= (B) Make adjustments to the weights

 

2. (True/False) The “vanishing gradient” problem can be solved using a different activation function

A) True
B) False

ANSWER= (A) True

 

3. (True/False) Every node in a neural network has an activation function.

A) True
B) False

ANSWER= (B) False

 

4. These are all activation functions except:

A) Sigmoid
B) Hyperbolic tangent
C) Leaky hyperbolic tangent
D) ReLu

ANSWER= (C) Leaky hyperbolic tangent

 

5. Deep Learning uses deep Neural Networks for all these uses, except

A) As an alternative to manual feature engineering
B) To uncover usually unobserved relationships in the data
C) Cases in which explainability is the main objective
D) As a classification and regression technique

ANSWER= (C) Cases in which explainability is the main objective

 

6. These are all activation functions except:

A) Tanh (hyperbolic tangent)
B) ReLU (Rectified Linear Unit
C) Softmax
D) Pruning

ANSWER= (D) Pruning

 

7. (True/False) Optimizer approaches for Deep Learning Regularization use gradient descent:

A) True
B) False

ANSWER= (B) False

 

8. Stochastic gradient descent is this type of batching method:

A) online learning
B) mini batch
C) full batch
D) stochastic batch

ANSWER= (A) online learning

 

9. (True/False) The main purpose of data shuffling during the training of a Neural Network is to aid convergence and use the data in a different order each epoch.

A) True
B) False

ANSWER= (A) True

 

10. Which of the following IS NOT a benefit of Transfer Learning?

A) Reducing time required to tune hyper-parameters
B) Reducing the impact of the vanishing gradient problem on early layers
C) Improving the speed at which large models can be trained from scratch
D) Conveying computational benefits when problems share similar primitive features.

ANSWER= (C) Improving the speed at which large models can be trained from scratch

 

14. Which of the following statements about using a Pooling Layer is TRUE?

A) Pooling can reduce both computational complexity and overfitting.
B) Pooling can reduce computational complexity, at the cost of overfitting.
C) Pooling increases computational complexity, but helps with overfitting.
D) Pooling reduces the likelihood of overfitting, but generally does not impact computational complexity.

ANSWER= (A) Pooling can reduce both computational complexity and overfitting.

Deep Learning and Reinforcement Learning Week 3 All Quiz

Practice: Optimizers and Data Shuffling 

1. True/False. Multi-layer perceptrons always have a hidden layer.

A) True
B) False

ANSWER= (A) True

 

2. True/False. Multi-layer perceptrons are considered a type of feedforward neural network.

A) True
B) False

ANSWER= (A) True

 

3. Select the correct rule of thumb regarding training a neural network. In general, as you train a neural network:

A) The log loss decreases and the accuracy decreases
B) The log loss decreases and the accuracy increases
C) The log loss increases and the accuracy decreases
D) The log loss increases and the accuracy increasess

ANSWER= (B) The log loss decreases and the accuracy increases

Week 3 Final Quiz  

1. What is the main function of backpropagation when training a Neural Network?

A) Preprocess the input layer
B) Make adjustments to the weights
C) Make adjustments to the loss function
D) Propagate the output on the output layer

ANSWER= (B) Make adjustments to the weights

 

2. (True/False) The “vanishing gradient” problem can be solved using a different activation function

A) True
B) False

ANSWER= (A) True

 

3. (True/False) Every node in a neural network has an activation function.

A) True
B) False

ANSWER= (B) False

 

4. These are all activation functions except:

A) Sigmoid
B) Hyperbolic tangent
C) Leaky hyperbolic tangent
D) ReLu

ANSWER= (C) Leaky hyperbolic tangent

 

5. Deep Learning uses deep Neural Networks for all these uses, except

A) As an alternative to manual feature engineering
B) To uncover usually unobserved relationships in the data
C) Cases in which explainability is the main objective
D) As a classification and regression technique

ANSWER= (C) Cases in which explainability is the main objective

 

6. These are all activation functions except:

A) Tanh (hyperbolic tangent)
B) ReLU (Rectified Linear Unit
C) Softmax
D) Pruning

ANSWER= (D) Pruning

 

7. (True/False) Optimizer approaches for Deep Learning Regularization use gradient descent:

A) True
B) False

ANSWER= (B) False

 

8. Stochastic gradient descent is this type of batching method:

A) online learning
B) mini batch
C) full batch
D) stochastic batch

ANSWER= (A) online learning

 

9. (True/False) The main purpose of data shuffling during the training of a Neural Network is to aid convergence and use the data in a different order each epoch.

A) True
B) False

ANSWER= (A) True

 

10. This is a high-level library that is commonly used to train deep learning models and runs on either TensorFlow or Theano:

A) PyTorch
B) Keras
C) Watson Studio
D) Deep Learning

ANSWER= (B) Keras

Deep Learning and Reinforcement Learning Week 2 All Quiz

Practice: Back Propagation, Activation Functions 

1. Select the method or methods that best help you find the same results as using matrix linear algebra to solve the equation θ=(XTX)−1XTy

A) Use stochastic gradient descent
B) Use scikit-learn to build a linear regression model
C) Train a neural network model
D) All the above

ANSWER= (D) All the above

 

2. (True/False) Neurons can be used as logic gates

A) True
B) False

ANSWER= (A) True

 

3. (True/False) The feed-forward computation of a neural network can be thought of as matrix calculations and activation functions.

A) True
B) False

ANSWER= (A) True

Practice: Keras Library 

1. Building a Neural Network with the Sequential API in Keras implies that each layer

A) can connect only to subsequent layers.
B) after the first layer must be fully-connected.
C) can connect to only the previous and next layers.
D) can be connected to at most two subsequent layers.

ANSWER= (C) can connect to only the previous and next layers.

 

2. An epoch in estimating a Deep Learning model refers to

A) subsamples that make up the training data.
B) iterations required to achieve convergence.
C) computational complexity of the estimating procedure.
D) the number of times the entire input data set is used by the model.

ANSWER= (D) the number of times the entire input data set is used by the model.

 

3. An advantage of the Sigmoid activation function over the step activation function is:

A) the ability to generate nonlinear outcomes.
B) sharper changes as the loss becomes positive.
C) improved backpropagation due to nonzero gradients.
D) significantly different gradients for very high values.

ANSWER= (C) improved backpropagation due to nonzero gradients.

Week 2 Final Quiz  

1. The backpropagation algorithm updates which of the following?

A) The parameters only.
B) The losses only.
C) The parameters and activations.
D) The activations only.

ANSWER= (A) The parameters only.

 

2. What of the following about the activation functions is true?

A) They add non-linearity into the model, allowing the model to learn complex pattern.
B) They are optimization algorithms that update values of the model parameters.
C) They evaluate how well the model has performed on the training data.
D) They tell us about how computationally expensive a neural network is.

ANSWER= (A) They add non-linearity into the model, allowing the model to learn complex pattern.

 

3. What is true regarding the backpropagation rule?

A) It prevents overfitting
B) The actual output is determined by computing the output of neurons in each hidden layer
C) It can be used to update the hyperparameters of a neural network
D) It is a feed forward neural network.

ANSWER= (B) The actual output is determined by computing the output of neurons in each hidden layer

 

4. Which option correctly lists the steps to build a linear regression model using Keras?
1. Use `fit()` and specify the number of epochs to train the model for.
2. Create a Sequential model with the relevant layers.
3. Normalize the features with ` layers.Normalization()` and apply `adapt()`.
4. Compile using `model.compile()` with specified optimizer and loss.

A) 3, 2, 4, 1
B) 3, 1, 2, 4
C) 3, 2, 1, 4
D) 2, 4, 3, 1

ANSWER= (A) 3, 2, 4, 1

 

5. (True/False) Keras provides one approach to build a model: by defining a Sequential model.

A) True
B) False

ANSWER= (B) False

Deep Learning and Reinforcement Learning Week 1 All Quiz

Practice: Introduction to Neural Networks 

1. Neural networks and Deep Learning are behind many of the AI applications that are part of our daily lives.

A) True
B) False

ANSWER= (A) True

 

2. Which one of the following is true in terms of the difference between grid search and randomized search?

A) Grid search is more efficient than randomized search.
B) The points in the randomized search space are more evenly distributed than the points in the grid search space.
C) Randomized search selects random combinations of parameters to train a model, whereas grid search goes through all combinations.
D) Randomized search goes through a more exhaustive search for selecting a model than grid search.

ANSWER= (C) Randomized search selects random combinations of parameters to train a model, whereas grid search goes through all combinations.

 

3. This is a characteristic that neural networks and logistic regression have in common:

A) both models retain easy explainability for their computational outcomes.
B) both models use only linear functions.
C) the weights, inputs, and bias of neural networks are the equivalent to the coefficients, variables, and constant of a logistic regression
D) both models make use of layers of units of computation.

ANSWER= (C) the weights, inputs, and bias of neural networks are the equivalent to the coefficients, variables, and constant of a logistic regression

Practice: Optimization and Gradient Descent 

1. Select all the methods that can be used to minimize a cost function:

A) mini-batch gradient descent
B) stochastic gradient descent
C) batch gradient descent

ANSWER= ( A) mini-batch gradient descent
(B) stochastic gradient descent
(C) batch gradient descent

 

2. How many sample(s) are used in a stochastic gradient descent?

A) 1
B) 2
C) 3
D) 4

ANSWER= (A) 1

 

3. Which method uses all the samples in one iteration to update model parameters?

A) Batch gradient descent
B) Stochastic gradient descent
C) Mini-batch gradient descent

ANSWER= (A) Batch gradient descent

Week 1 Finall Quiz  

1. What is another name for the “neuron” on which all neural networks are based?

A) deep neuron
B) sigmoid
C) neutron
D) perceptron

ANSWER= (D) perceptron

 

2. What is an advantage of using a network of neurons?

A) The network is not limited to using only the sigmoid function as an activation function.
B) A network of neurons can represent a non-linear decision boundary.
C) Feedforward capabilities are limited.
D) The output of neurons can be averaged.

ANSWER= (B) A network of neurons can represent a non-linear decision boundary.

 

3. A dataset with 8 features would have how many nodes in the input layer?

A) 10
B) 2
C) 4
D) 8

ANSWER= (D) 8

 

4. For a single data point, the weights between an input layer with 3 nodes and a hidden layer with 4 nodes can be represented by a:

A) 4 x 3 matrix
B) 3 x 4 matrix.
C) 4 x 4 matrix
D) 3 x 3 matrix

ANSWER= (B) 3 x 4 matrix.

 

5. Use the following image for reference. How many hidden layers are in this Neural Network?

A) Two
B) Four
C) Eight
D) Fourteen

ANSWER= (A) Two

 

6. Use the following image for reference. How many hidden units are in this Neural Network?

A) Two
B) Four
C) Eight
D) Fourteen

ANSWER= (C) Eight

 

7. Which statement is TRUE about the relationship between Neural Networks and Logistic Regression?

A) A Neural Network is less likely to overfit to training data than Logistic Regression.
B) A Neural Network with two or more deep layers will likely outperform Logistic Regression.
C) A Multi-Layer Perceptron is equivalent to Logistic Regression if all activation functions are the same.
D) A single-layer Neural Network can be parameterized to generate results equivalent to Linear or Logistic Regression.

ANSWER= (D) A single-layer Neural Network can be parameterized to generate results equivalent to Linear or Logistic Regression.