Dashbo… / My cour… / Electronics and Telecommunication Engin… / EXTC-Aug23-Feb24 Co… / D… / 11 September - 17 Septe… / TES…

Started on Sunday, 17 September 2023, 8:33 AM

State Finished
Completed on Sunday, 17 September 2023, 8:53 AM
Time taken 20 mins 1 sec
Grade 17.00 out of 20.00 (85%)

Question 1

Correct

Mark 1.00 out of 1.00

The complexity in ANN depends upon the type of functional mapping, accurate and sufficient training data acquired, their effective way of
representation. Weighted sum in ANN are referred to as

a. Activation

b. Link

c. output

d. Input

The correct answer is: Activation

Question 2
Correct

Mark 1.00 out of 1.00

Artificial Neural Network (ANN) are type of Machine Learning model that are inspired by the structure and function of the human brain. In
________________ type of ANN information passed till it reaches the output.

a. Convolutional Neural Networks

b. Recurrent Neural Networks

c. Feed Forward Neural Networks

d. Deconvolutional Neural Networks

The correct answer is: Feed Forward Neural Networks

Question 3

Incorrect

Mark 0.00 out of 1.00

ANN used for Classification, Clustering and Pattern recognition. A single layer feedforward neural network with preprocessing is perceptron.
If the change in weight vector is represented by Δ w ij what does it mean?

a. Describes the change in weight vector for ith processing unit, taking input vector jth into account

b. Describes the change in weight vector for jth processing unit, taking input vector ith into account 

c. Describes the change in weight vector for jth and ith processing unit

d. None of the mentioned

The correct answer is: Describes the change in weight vector for ith processing unit, taking input vector jth into account

Question 4
Correct

Mark 1.00 out of 1.00

Perceptron learning rule states that the algorithm would automatically learn the optimal weight coefficient. To find minimum or maximum
of a function, we set gradient to zero because which of the follwing

a. Depends on type of problem

b. None of all

c. both the conditions must satisfied

d. the value of gradient at extrema of a function is always zero

The correct answer is: the value of gradient at extrema of a function is always zero

Question 5

Correct

Mark 1.00 out of 1.00

Which of the following true for neural networks?

A. The training time depends on size of the network

B. Neural networks can be simulated on a conventional computer.

c. Artificial neurons are identical in operation to biological ones.

a. NONE

b. ALL

c. A AND B

d. ONLY B

The correct answer is: A AND B

Question 6

Correct

Mark 1.00 out of 1.00

The back propagation law is also known as generalized delta rule. Error in output is propagated backwards only to determine weight
updates. There is no feedback of signal at any stage. The objective of back propagation algorithm is _______________

a. To develop learning algorithm for multilayer feedforward neural network, so that network can be trained to capture the 
mapping implicitly

b. To develop learning algorithm for multilayer feedforward neural network

c. None of the mentioned

d. To develop learning algorithm for single layer feedforward neural network

The correct answer is: To develop learning algorithm for multilayer feedforward neural network, so that network can be trained to capture
the mapping implicitly

Question 7
Correct

Mark 1.00 out of 1.00

A Gradient Descent algorithm is the key to minimizing the loss function and achieving our target, which is to predict close to the original
value. Does the back propagation learning is based on Gradient Descent algorithm along error surface

a. YES

b. NO

c. CAN'T BE SAID

d. It depends on Gradient descent but not along error surface.

The correct answer is: YES

Question 8
Correct

Mark 1.00 out of 1.00

Outliers are those data points that are significantly different from the rest of the dataset. Which statement about outliers is true?

a. Outliers should be identified and removed from a dataset.

b. The nature of the problem determines how outliers are used.

c. Outliers should be part of of the test dataset but should not be present in the training dataset.

d. Outliers should be part of the training dataset but should not be present in the test data set

The correct answer is: The nature of the problem determines how outliers are used.
Question 9

Correct

Mark 1.00 out of 1.00

Supervised learning is the type of machine learning that uses labeled data to train a model to make predictions. unsupervised learning on
the other hand, doesn't require labeled data and instead uses algorithms to find patterns and relationships in the data. What is the
purpose of regularization in deep learning?

a. To increase the accuracy of the model.

b. To reduce the number of features in a model.

c. To prevent overfitting and improve generalization.

d. To speedup training process.

The correct answer is: To prevent overfitting and improve generalization.

Question 10
Correct

Mark 1.00 out of 1.00

A validation data set and test data set are not the same thing. A validation data set is usually necessary in machine/ deep learning to
prevent overfitting. A validation data set is used to___________ while a test set is used to ____________________.

a. A validation set is used to evaluate the performance of a model during training, while a test set is used to evaluate its 
performance after training.

b. Both statements are wrong

c. Both statements are correct

d. A validation set is used to tune the hyperparameters of a mode, while a test set is used to evaluate its performance

The correct answer is: A validation set is used to evaluate the performance of a model during training, while a test set is used to evaluate its
performance after training.

Question 11

Incorrect

Mark 0.00 out of 1.00

The Akaike information criterion is a measure of model complexity used in machine /deep learning and statistics to compare the
performance of different models. It takes into account both the goodness of fit and the number of parameters in the model, and penalizes
models with more parameters. What is the purpose of data augmentation in deep learning?

a. to improve the performance of a model 

b. To improve the quality of a data set

c. to decrease size of data set

d. To increase size of data set

The correct answer is: To increase size of data set

Question 12

Correct

Mark 1.00 out of 1.00

A hyperparameter is a parameter that is set before the training process begins and cannot be learned directly from the data. Examples of
hyperparameters include________ which determines the step size taken during gradient descent optimization and the number of hidden
units in a neural network which controls its capacity and complexity.

a. Area Under the ROC (AUC)

b. Mean Squared Error (MSE)

c. Accuracy

d. Learning rate

The correct answer is: Learning rate

Question 13
Correct

Mark 1.00 out of 1.00

Non-linear relationships are common problems in regression tasks, where response variable does not vary linearly with the predictors. There
are several approaches to resolve these problems. Which of the following is the common approach to solve regression problem with
nonlinear relationships?

a. Polynomial regression

b. Linear Regression

c. Neural Network

d. all of them

The correct answer is: Neural Network

Question 14
Correct

Mark 1.00 out of 1.00

In PCA the number of input dimension are equal to principal components. PCA can be used for projecting and visualizing data in lower
dimensions. In following ___________________type of feature selection method we start with empty feature set

a. None of them

b. In both

c. Backward feature selection

d. forward feature selection

The correct answer is: forward feature selection

Question 15

Incorrect

Mark 0.00 out of 1.00

The PCA method is used for feature extraction. The output of training process in machine learning is a machine learning model.
Dimensionality reduction algorithms are one of the possible ways to reduce the computation time required to build a model. Which of the
following techniques would perform better for dimensions of a data set?

a. Removing columns which have high variance in data. 

b. None of these

c. Removing columns which have too many missing values.

d. Removing columns which have dissimilar data trends

The correct answer is: Removing columns which have too many missing values.

Question 16
Correct

Mark 1.00 out of 1.00

Penalized logistic regression imposes a penalty to the logistic model for having too many variables. This results in shrinking the coefficient
of the less contributive variables towards___________

a. zero

b. one

c. Infinity

d. Nonnumeric constant value

The correct answer is: zero

Question 17
Correct

Mark 1.00 out of 1.00

Ensemble methods are techniques that create multiple models and then combine them to produce improved results. Ensemble learning
techniques employ ______________________________________

a. Reduce accuracy of classification models or to increase mean absolute error for regression models.

b. To increase the accuracy of classification models or to reduce mean absolute error for regression models. 

c. To increase the accuracy of classification models or to increase mean absolute error for regression models.

d. To reduce the accuracy of classification models or to reduce absolute error for regression models.

The correct answer is: To increase the accuracy of classification models or to reduce mean absolute error for regression models.
Question 18

Correct

Mark 1.00 out of 1.00

Current Neural Network based feature selection methods employs a simple autoencoder to perform feature selection based on
reconstruction error. For feature selection it is necessary to have a simple network so that error can be easily back propagated. Typically
Feature selection method consist of

1. Subset Generation
2. Subset evaluation

3. stopping criteria

4. Result validation

a. only two required

b. only three steps required

c. All four

d. No steps required

The correct answer is: All four

Question 19

Correct

Mark 1.00 out of 1.00

In deep neural network, what is the role of nonlinear activation functions such as sigmoid, Tanh and ReLU?

a. They help to learn nonlinear decision boundaries.

b. They unlearn nonlinear behaviors.

c. They speed up gradient calculations in back propagation as compared to linear units.

d. They can introduce bias.

The correct answer is: They help to learn nonlinear decision boundaries.
Question 20

Correct

Mark 1.00 out of 1.00

Classification is a type of supervised learning where a target feature, which is of categorial type, is predicted for test data on the basis of
information imparted by the training data. The target categorial feature is known as

a. variable

b. object

c. class

d. method

The correct answer is: class

◄ lab assignment 3

Jump to...

book ch 9 CNN pdf ►

Dashb… / My c… / Electronics and Telecommunicati… / EXTC-Aug23-Feb2… / D… / 16 October - 22… / TEST2 DEEP LEARNING AND EDG…

Started on Sunday, 22 October 2023, 11:38 AM

State Finished
Completed on Sunday, 22 October 2023, 11:58 AM
Time taken 20 mins
Grade 12.00 out of 20.00 (60%)

Question 1

Not answered

Marked out of 1.00

Autoencoders can be viewed as a generalization of PCA.

A. One way to implement undercomplete autoencoder is to constrain the number of nodes present in hidden layers of the neural network.
B. Autoencoders are capable of learning a continuous nonintersecting surface.

a. Statement 1 is false and 2 is true.

b. Both statements are true.

c. Statement 1 is true and 2 is false.

d. Both statements are false.

The correct answer is: Both statements are true.

Question 2
Correct

Mark 2.00 out of 2.00

Suppose you have built a neural network. You decide to initialize the weights and biases to be zero. Which of the following statements are
true?

1. Each neuron in the first hidden layer will perform the same computation. So even after multiple iterations of gradient descent each
neuron in the layer will be computing the same thing as other neurons.

2. Each neuron in the hidden layer will perform the same computation in the first iteration. But after one iteration of gradient descent they
will learn to compute different things because we have broken symmetry.

3. Each neuron in the first hidden layer will compute the same thing, but neurons in different layers will compute different things.

4. The first hidden layers neurons will perform different computations from each other even in the first iteration; their parameters will thus
keep evolving in their own way.

a. Statement 3 is true.

b. Statement 2 is true.

c. Statement 4 is true.

d. Statement 1 is true.

The correct answer is: Statement 1 is true.

Question 3

Correct

Mark 1.00 out of 1.00

What is the basic concept of Recurrent Neural Network?

a. Use recurrent features from dataset to find the best answers.

b. Use pervious inputs to find the next output according to the training set.

c. Use a loop between inputs and outputs in order to achieve the better prediction.

d. Use loops between the most important features to predict next output.

The correct answer is: Use pervious inputs to find the next output according to the training set.

Question 4

Correct

Mark 1.00 out of 1.00

Padding on the input volume with zeros in such a way that the convolution layer does not alter the spatial dimensions of the input

a. Keep the spatial dimensions of input intact.

b. Alter the spatial dimensions of the input.

c. All are incorrect...

d. Keep the output spatial dimensions intact.

The correct answer is: Keep the spatial dimensions of input intact.

Question 5
Incorrect

Mark 0.00 out of 1.00

A. Autoencoders are unsupervised learning techniques.

B. Autoencoders output is same as the input.

a. Both statements are true.

b. Statement A is false. Statement B is true.

c. Both statements are false.

d. Statement A is true. Statement B is false. 

The correct answer is: Both statements are true.

Question 6

Correct

Mark 1.00 out of 1.00

You are training an RNN and find that you are weights and activation are all taking on the value of " not a number". Which of these is the
most likely cause of this problem?

a. ReLU activation function used to compute with too large input value.

b. Sigmoid activation function used to compute with too large input value.

c. Exploding gradient

d. Vanishing gradient

The correct answer is: Exploding gradient

Question 7
Incorrect

Mark 0.00 out of 1.00

Benefits of Pooling

1. Reduces dimensions and computation

2. Reduce overfitting as there are less parameters

3. Model is intolerant towards variations and distortions

a. Point 1 and 3 true. 

b. Point 2 and 3 true.

c. Point 1 and 2 true.

d. All points are true

The correct answer is: Point 1 and 2 true.

Question 8

Incorrect

Mark 0.00 out of 1.00

Which of the following is False about LSTM?

a. LSTM is an extension for RNA which extends its memory. 

b. LSTM solves the exploding gradients issues in RNN.

c. LSTM enables RNN to learn long-term dependencies.

d. None of the above.

The correct answer is: LSTM solves the exploding gradients issues in RNN.
Question 9

Correct

Mark 1.00 out of 1.00

For a Input image 6*6, stride size 1 and convolution filter size 3*3....... output Feature map size is

a. None of the all

b. 3*3

c. 4*4

d. 6*6

The correct answer is: 4*4

Question 10

Incorrect

Mark 0.00 out of 1.00

GATED RECURRENT UNIT (GRU) network is one of the most famous alternative architectures.

1. It combines the forget and input gates into a single update gate.

2. it also merges the cell state and hidden state and makes some other changes.

3. Generates an output at each time step and this output used to train the network

a. All statements are false. 

b. Statement 2 and 3 are true.

c. Statement 1 and 2 are true.

d. statement 1 and 3 are true.

e. All statements are true.

The correct answer is: All statements are true.

Question 11
Correct

Mark 1.00 out of 1.00

Which of the following activation functions can lead to Vanishing gradients?

a. Leaky ReLU

b. None of the above

c. Tanh

d. ReLU

The correct answer is: Tanh

Question 12

Incorrect

Mark 0.00 out of 1.00

Convolution leverages three important ideas that can help improve a Machine/ Deep learning system

A. Sparsh Interactions B. Parameter Sharing C. Equivariant Representations.

The parameter sharing used by the convolution operation menace that rather than learning a separate set of parameters for every location
we learn only one set.

1. This does not affect the runtime of forward propagation.

2. It does not reduce the storage requirements of the model to k parameters.

3. It does reduce the storage requirements of the model to k parameters.

a. 1st and 2nd points are true. 

b. 2nd and 3rd points are true.

c. 1st and 3rd points are true.

d. All points are true.

The correct answer is: 1st and 3rd points are true.

Question 13
Correct

Mark 1.00 out of 1.00

We can reduce the exploding gradients problem by clipping the gradient during back propagation so that they never exceed some
threshold. This is called___________

a. Gradient Vanishing

b. Gradient Clipping

c. Gradient Overacting

d. Gradient Cutting

The correct answer is: Gradient Clipping

Question 14

Correct

Mark 1.00 out of 1.00

The Gated Recurrent Unit (GRU) cell was proposed by Kyunghyun Cho et.al. in the year _________

a. 2002

b. 2016

c. 2014

d. 2019

The correct answer is: 2014

Question 15

Incorrect

Mark 0.00 out of 1.00

The state of a cell in LSTM at time t is Ct.

1. The LSTM modify the state only through linear interactions: information flows smoothly across time.

2. The LSTM modify the state only through nonlinear interactions: information flows randomly across time.

3. LSTM protect and control the information in the cell through gates.
4. LSTM gates are implemented by tanh and a pointwise multiplication.

a. statement 1 and 4 are true. 

b. statement 2 and 3 are true.

c. Statement 1 and 3 are true.

d. statement 2 and 4 are true.

e. Statement 1 and 2 are true.

The correct answer is: Statement 1 and 3 are true.

Question 16

Correct

Mark 1.00 out of 1.00

The denoising encourages the encoder to keep important information but forget about spurious information about the input.

A. The loss is between the original input and the reconstruction from a noisy version of the input.

B. Denoising autoencoders can be used as a tool for feature extraction.

a. Both statements are true.

b. Only statement 2 is true.

c. Only Statement 1 is true.

d. Both statements are false.

The correct answer is: Both statements are true.

Question 17
Incorrect

Mark 0.00 out of 1.00

In RNN each unit has an internal state which is called the_________________________________

a. visible state of unit

b. hidden function 

c. visible function

d. hidden state of unit

The correct answer is: hidden state of unit

Question 18
Correct

Mark 1.00 out of 1.00

You are building a binary classifier for recognizing cucumbers (y=1) Vs watermelons (y=0). Which one of these activation functions would
you recommend using for the output layers

a. Sigmoid

b. Leaky ReLU

c. tanh

d. ReLU

The correct answer is: Sigmoid

Question 19

Correct

Mark 1.00 out of 1.00

Which of the following is FALSE about LSTM?

a. These gates are digital in nature.

b. LSTM stores information in these gated cells.

c. Data can be stored, deleted and read from these gated cells like computer storage.

d. Gates of these cells open and closed based on some decisions.

The correct answer is: These gates are digital in nature.

◄ CNN

Jump to...