Download ARPIT 2019 Neural Networks And Deep Learning Creation Question Paper

Question Paper Name: 5324 Neural Networks and Deep Learning 30th June 2019 Shift 1

Subject Name: Neural Networks and Deep Learning

Creation Date: 2019-06-30 13:01:46

--- Content provided by‍ FirstRanker.com ---

Duration: 180

Total Marks: 100

Display Marks: Yes

Neural Networks and Deep Learning

Group Number: 1

--- Content provided by​ FirstRanker.com ---

Group Id: 489994230

Group Maximum Duration: 0

Group Minimum Duration: 120

Revisit allowed for view?: No

Revisit allowed for edit?: No

--- Content provided by FirstRanker.com ---

Break time: 0

Group Marks: 100

Neural Networks and Deep Learning

Section Id: 489994286

Section Number: 1

--- Content provided by​ FirstRanker.com ---

Section type: Online

Mandatory or Optional: Mandatory

Number of Questions: 29

Number of Questions to be attempted: 29

Section Marks: 100

--- Content provided by‌ FirstRanker.com ---

Display Number Panel: Yes

Group All Questions: No

Sub-Section Number: 1

Sub-Section Id: 489994312

Question Shuffling Allowed: Yes

--- Content provided by FirstRanker.com ---

Question Number: 1 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

What is the secret of success of deep learning?

1. Vanishing gradient problem does not exist anymore
2. Number of weights associated with the network have become in the order of billions

--- Content provided by‌ FirstRanker.com ---

3. Brute computing power using GPUs have become manifold
4. Machine intelligence has almost become equivalent to human intelligence

Options:

1.1 www.FirstRanker.com

Question Number: 2 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

--- Content provided by‍ FirstRanker.com ---

Correct Marks: 1 Wrong Marks: 0

When is the cell said to be fired?

1. If potential of the cell body reaches a steady threshold value
2. If there is impulse reaction
3. During upbeat of heart

--- Content provided by FirstRanker.com ---

4. None of the above

Options:

1.1

2.2

3.3

--- Content provided by‌ FirstRanker.com ---

4.4

Question Number: 3 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

Who developed the first learning machine in which connection strengths could be adapted automatically?

1. McCulloch-pits

--- Content provided by​ FirstRanker.com ---

2. Marvin Minsky
3. Hopfield
4. Rosenblatt

Options:

1.1

--- Content provided by​ FirstRanker.com ---

2.2

3.3

4.4

Question Number: 4 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

--- Content provided by‌ FirstRanker.com ---

Who proposed the first perceptron model in 1958?

1. McCulloch-pits
2. Marvin Minsky
3. Hopfield
4. Rosenblatt

--- Content provided by FirstRanker.com ---

Options:

1.1

2.2

3.3

4.4

--- Content provided by FirstRanker.com ---

Question Number: 5 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

sign of weight indicates?

1. Excitatory input
2. Inhibitory input

--- Content provided by‌ FirstRanker.com ---

3. Can be either excitatory or inhibitory as such
4. None of the Above

Options:

1.1

2.2

--- Content provided by⁠ FirstRanker.com ---

3.3

4.4

Question Number: 6 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

Competitive learning net is used for?

--- Content provided by FirstRanker.com ---

1. Pattern grouping
2. Pattern storage
3. Pattern grouping or storage
4. None of the above

Options:

--- Content provided by‍ FirstRanker.com ---

1.1

2.2

3.3

4.4

Question Number: 7 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

--- Content provided by FirstRanker.com ---

Correct Marks: 1 Wrong Marks: 0

Which loss function out of the following is best for image classification (one class out of many) using deep neural network?

1. Categorical cross entropy
2. Mean squared error loss
3. Binary cross entropy

--- Content provided by⁠ FirstRanker.com ---

4. Hinge loss

Options:

1.1

2.2

3.3

--- Content provided by‌ FirstRanker.com ---

4.4

Question Number: 8 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

The problem that a CNN with 50 layers performs worse than a CNN with 30 layers on both training and test data can be effectively resolved by using

1. Regularization

--- Content provided by⁠ FirstRanker.com ---

2. Validation set
3. Resnet architecture
4. Softmax activation

Options:

1.1

--- Content provided by​ FirstRanker.com ---

2.2

3.3

4.4 www.FirstRanker.com

Question Number: 9 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

--- Content provided by‌ FirstRanker.com ---

In Delta Rule for mean squared error minimization, weights are adjusted proportional to

1. the changes in the output vector
2. the difference between desired output and actual output
3. the difference between input and output of the multi-layered perceptron
4. None of the above

--- Content provided by⁠ FirstRanker.com ---

Options:

1.1

2.2

3.3

4.4

--- Content provided by‌ FirstRanker.com ---

Question Number: 10 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

What is a "hidden layer" of a neural network?

1. The last layer of neurons, which is hidden from the inputs.
2. One of the middle layers, which aren't directly connected to either inputs or outputs.

--- Content provided by​ FirstRanker.com ---

3. A group of neurons with zero weights, which have effectively hidden themselves from the rest of the network.
4. A layer of neurons that's in control of turning inputs on or off, which affects how training is accomplished.

Options:

1.1

2.2

--- Content provided by​ FirstRanker.com ---

3.3

4.4

Question Number: 11 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

Given variables: (i) the present input; (ii) the previous cell state; (iii) the previous output; (iv) the previous hidden state; the response of a traditional LSTM depends on:

--- Content provided by⁠ FirstRanker.com ---

1. (i), (ii), (iii)
2. (i), (ii), (iii),(iv)
3. (i), (ii),(iv)
4. (i), (iii), (iv)

Options:

--- Content provided by⁠ FirstRanker.com ---

1.1

2.2

3.3

4.4 www.FirstRanker.com

Question Number: 12 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

--- Content provided by FirstRanker.com ---

Correct Marks: 1 Wrong Marks: 0

Which of the following is correct regarding Gated Recurrent Units (GRUs)?

1. It combines the forget and input gates into a single "update gate."
2. It merges the cell state and hidden state
3. It takes the previous hidden state multiplied by reset gate as an input.

--- Content provided by⁠ FirstRanker.com ---

4. All of these

Options:

1.1

2.2

3.3

--- Content provided by​ FirstRanker.com ---

4.4

Question Number: 13 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

Please mark the correct weight update law for RBM. Here, W stands for weights of RBM, b stands for hidden unit bias vector of RBM, c stands for visible layer bias vector of RBM and ? = 1 is the learning rate.

1. W = W + ?(h(x^(t))x^(t) - h(x^(t))x^(t)); b = b + ?(h(x^(t)) - h(x^(t)); c = c + (x^(t) - x^(t)))

--- Content provided by⁠ FirstRanker.com ---

2. W = W - ?(h(x^(t))x^(t) - h(x^(t))x^(t)); b = b + ?(h(x^(t)) - h(x^(t)); c = c - ?(x^(t) - x^(t)))
3. W = W + ?(h(x^(t))x^(t) - h(x^(t))x^(t)); b = b - ?(h(x^(t)) - h(x^(t)); c = c + ?(x^(t) - x^(t)))
4. W = W - ?(h(x^(t))x^(t) - h(x^(t))x^(t)); b = b - ?(h(x^(t)) - h(x^(t)); c = c - (x^(t) - x^(t)))

Options:

1.1

--- Content provided by⁠ FirstRanker.com ---

2.2

3.3

4.4

Question Number: 14 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

--- Content provided by‍ FirstRanker.com ---

Which deep network was used for the image classification for the first time in image-net challenge?

1. VGG
2. Googlenet
3. ZFnet
4. Alexnet

--- Content provided by‍ FirstRanker.com ---

Options:

1.1

2.2

3.3

4.4 www.FirstRanker.com

--- Content provided by​ FirstRanker.com ---

Question Number: 15 Question Type: MCQ Option Shuffling: No Display Question Number: Yes Single Line Question Option: No Option Orientation: Vertical

Correct Marks: 1 Wrong Marks: 0

Which deep network was used for the image classification for the first time in image-net challenge?

1. VGG
2. Googlenet

--- Content provided by​ FirstRanker.com ---

3. ZFnet
4. Alexnet

In the following figure, please mark the correct expression for the back-error propagated for the 1st layer

This download link is referred from the post: ARPIT Last 10 Years 2011-2021 Previous Question Papers