Artificial Intelligence For Engineering Quizzes Part 8

Q:1. The incorrect statement for a Convolutional Neural Network are:

1.The height and width of the filter in CNN must be less than the size of input

2.The Pooling layer progressively increases the spatial size of the representation

3.It uses both linear and non-linear activation functions

4.The last few layers are fully connected layers and computation on these layers are very time consuming

Solution- 1.The height and width of the filter in CNN must be less than the size of input

Reason- This is because the height and width of the filter in CNN must not be less than the size of input.

Q:2. A Convolutional Neural Network is able to successfully capture the Spatial and Temporal dependencies:

1.True

2.False

Solution- 1.True

Reason- Yes this true, because a ConvNet is able to successfully capture the Spatial and Temporal dependencies in an image through the application of relevant filters.

Q:3. Different types of normalization in Deep Neural Networks are

a. Output

b. Batch

c. Group

d. Instance

1.a,b,c

2.b,c,d

3.d,a,b

4.d,a,c

Solution- 2.b,c,d

Reason- Different types of normalization in Deep Neural Networks are batch, group, instance, layer, weight etc..

Q:4. Applications of CNNs are:

a. Recommender systems

b. AlexNet

c. Natural Language Processing

d. Pooling

1.a,b

2.b,d

3.a,c

4.a,d

Solution- 4.a,d

Reason- This is because recommender system is an application of CNN and Pooling layers are used to reduce the dimensions of the feature maps in CNN.

Q:5. Which of the following statements are correct for GAN?

a. GANs are useful for unsupervised learning, supervised learning, semi-supervised learning, and reinforcement learning

b. Generative model technique learns to generate the data with the same statistics of training data

c. At each iteration the goal of generator is to minimize the classification error and the goal of discriminator is to maximize the classification error.

d. The discriminator could tell the difference between images of a cat and a dog and generative model could generate new images of animals that look like real animals.

1.a,b,c

2.a,b,d

3.a,c,d

4.b,c,d

Solution- 2.a,b,d

Reason- All the three a, b and d options are true.

Q:6. A generative model:

a. Captures the joint probability p(X,Y)

b. Captures the conditional probability p(Y|X)

c. Includes the distribution of data itself

d. Cannot predict the next word in sequence

1.a,b

2.a,c

3.a,d

4.b,c

Solution- 2. a, c

Reason- A generative model includes the distribution of the data itself, and tells you how likely a given example is. For example, models that predict the next word in a sequence are typically generative models (usually much simpler than GANs) because they can assign a probability to a sequence of words. and it. It also captures the joint probability p(X,Y)

Q:7. The discriminative model:

a. Draw boundaries in the data space as it tells the difference between handwritten 0s and 1s.

b. Captures the joint probability p(X,Y)

c. Captures the conditional probability p(Y|X)

d. Learns to distinguish the generator’s fake data from real data

1.a,b,c

2.a,b,d

3.a,c,d

4.b,c,d

Solution- 3. a,c,d

Reason- The discriminative model tries to tell the difference between handwritten 0’s and 1’s by drawing a line in the data space. It captures the conditional probability p(Y|X) and Learns to distinguish the generator’s fake data from real data

Q:8. Choose the incorrect statements from the following

1.The discriminator uses the real data as negative examples during training

2.The discriminator uses the fake data as negative examples during training

3.The portion of the GAN that trains the generator model includes random input

4.None of the above

Solution- 1.The discriminator uses the real data as negative examples during training

Reason- The generated instances become negative training examples for the discriminator. The discriminator learns to distinguish the generator’s fake data from real data.

Q:9. Choose the correct statements from the following

a. Most universal approximation theorems can be parsed into two classes. The first quantifies the approximation capabilities of neural networks with an arbitrary number of artificial neurons and the second quantifies an arbitrary number of hidden layers

b. A neural network can represent any function provided it has sufficient capacity.

c. Not all architectures can represent any function

d. None of the above

1.a,b,c

2.b,c,d

3.d,a,b

4.d,a,c

Solution- 1.a,b,c

Reason- All the above options a,b and c are true. As Most of the universal approximation theorems can be parsed into two class. The first class quantifies the approximation capabilitiesof neural networks with an arbitrary number of artificial neurons and the second quantifies an arbitrary numberof hidden layers

Q:10. Interesting applications of Generative Adversarial Networks (GANs) are:

1.Photo Inpainting

2.Culinary arts (as making a pizza)

3.Face aging

4.All the above

Solution- 4. All the above

Reason- Photo Inpainting, Culinary arts (as making a pizza), Face aging all are the applications of Generative Adversarial Networks (GANs)

Disclaimer-

This article only attempts to discover some questions that can be generated in Artificial Intelligence For Engineering with the answer to all these questions. There can be some errors to these answers. If you find any error then please do write to us.