WorksheetsGAN_MCQ
Total questions: 50
Worksheet time: 38mins
Name
Class
Date
1.
Which is the basic building block of deep learning?
a)
Decision Tree
b)
Random Forest
c)
Artificial Neural Network
d)
Support Vector Machine
2.
What is a perceptron?
a)
A convolution layer
b)
A single-layer neural network
c)
A pooling layer
d)
A type of data input
3.
What does a neural network's "weight" represent?
a)
Processing speed
b)
Adjustable parameter influencing output
c)
Storage capacity
d)
None of the above
4.
Which activation function is commonly used in hidden layers?
a)
Sigmoid
b)
ReLU
c)
Tanh
d)
All of the above
5.
What is the main purpose of backpropagation in ANN?
a)
Data normalization
b)
Optimizing weights using error gradients
c)
Feature selection
d)
Input data encoding
6.
Which optimization algorithm is widely used for training neural networks?
a)
Gradient Descent
b)
Nearest Neighbor
c)
Apriori
d)
K-Means
7.
In deep learning, what is dropout used for?
a)
Increase model complexity
b)
Reduce overfitting
c)
Data cleaning
d)
Ensemble learning
8.
What is data normalization?
a)
Scaling input data to a fixed range
b)
Classifying data
c)
Encoding categorical features
d)
Removing duplicates
9.
How are neural networks commonly initialized?
a)
Zero weights
b)
One weights
c)
Small random numbers
d)
Large constant numbers
10.
Which method speeds up training by transforming distributions to have zero mean and unit variance?
a)
Data augmentation
b)
Batch normalization
c)
Pooling
d)
Early stopping
11.
Deep learning primarily refers to neural networks with:
a)
One layer
b)
Two layers
c)
Multiple layers
d)
No layers
12.
Which problem can be mitigated by using LSTM or GRU layers in RNNs?
a)
Model overfitting
b)
Vanishing gradient
c)
Data imbalance
d)
Feature redundancy
13.
What is the role of an activation function in a neural network?
a)
To introduce non-linearity
b)
To reduce input size
c)
To sort data
d)
To group outputs
14.
Which of the following is NOT a type of deep learning architecture?
a)
CNN
b)
RNN
c)
GAN
d)
K-Means
15.
Which of these is an example of supervised learning?
a)
Image classification with labeled data
b)
GAN training
c)
K-Means clustering
d)
PCA
16.
What are the main components of a GAN?
a)
Generator and Encoder
b)
Discriminator and Encoder
c)
Generator and Discriminator
d)
Encoder and Decoder
17.
The generator in a GAN aims to:
a)
Discriminate fake from real data
b)
Generate synthetic (fake) data
c)
Label input images
d)
None of the above
18.
The primary purpose of the discriminator network is to:
a)
Generate new data
b)
Distinguish real from fake data
c)
Cluster the data
d)
Reduce noise
19.
What is the training process in GANs called?
a)
Supervised learning
b)
Data augmentation
c)
Adversarial learning
d)
Reinforcement learning
20.
The "latent space" in a GAN is used to:
a)
Store real images
b)
Control diversity of generated data
c)
Evaluate the model
d)
Encode class labels
21.
Which value type is usually input to the generator in a GAN?
a)
Random noise vector
b)
Real image
c)
Class label
d)
Binary mask
22.
What happens if a GAN suffers from "mode collapse"?
a)
The discriminator outperforms
b)
The generator produces limited types of outputs
c)
Large memory usage
d)
Training never starts
23.
What loss function is traditionally used in vanilla GANs?
a)
Mean squared error
b)
Binary cross-entropy
c)
Categorical cross-entropy
d)
L1 loss
24.
What is the best approach to avoid overfitting in GANs?
a)
Use larger batch sizes
b)
Use more complex generator
c)
Data augmentation and regularization
d)
None of these
25.
What is feature matching in GANs?
a)
Matching the distribution statistics of real and fake features
b)
Matching the generator and discriminator architectures
c)
Matching learning rates
d)
None of the above
26.
Which is a major challenge in GANs?
a)
Fast convergence
b)
Mode collapse
c)
Low computational cost
d)
Large labeled datasets
27.
Why is GAN training often unstable?
a)
No gradient flow to generator
b)
Discriminator easily outperforms generator
c)
Imbalanced loss optimization
d)
All of the above
28.
Which technique stabilizes GAN training by penalizing the norm of the gradient?
a)
Batch normalization
b)
Gradient penalty
c)
Early stopping
d)
Feature selection
29.
Which loss is minimized by the generator in standard GANs?
a)
log(1 - D(G(z)))
b)
log(D(G(z)))
c)
Mean squared error
d)
L2 loss
30.
When does a GAN reach equilibrium?
a)
Generator loss is zero
b)
Discriminator is always correct
c)
Both networks cannot improve further and output a Nash Equilibrium
d)
Discriminator loss is always lower than the generator
31.
DCGANs replace fully connected layers with:
a)
Pooling layers
b)
Convolutional layers
c)
Normalization layers
d)
Dropout layers
32.
Which dataset is commonly used for training DCGANs?
a)
ImageNet
b)
CelebA
c)
IMDB
d)
Wikidata
33.
DCGANs primarily improve GANs for which application?
a)
Text summarization
b)
Image generation
c)
Video ranking
d)
Sentiment analysis
34.
What main activation does DCGAN use in hidden layers?
a)
ReLU
b)
Sigmoid
c)
Tanh
d)
Softmax
35.
What is a key architectural innovation in DCGAN?
a)
Use of pooling layers
b)
Strided convolutions instead of pooling
c)
Recursive NN blocks
d)
LSTM units
36.
Conditional GANs are different from vanilla GANs because:
a)
Both networks receive a conditional input
b)
Only the generator receives random noise
c)
Outputs text data only
d)
Have no generator
37.
cGANs can be used for:
a)
Generating labeled images
b)
Dimensionality reduction
c)
Data cleaning
d)
Sequence prediction
38.
The label passed to cGANs is:
a)
Ignored by the generator
b)
Used as a condition for both generator and discriminator
c)
Post-processed outside the network
d)
Not used in training
39.
cGAN loss function includes which component?
a)
Regularization term for labels
b)
Conditional loss
c)
Only binary cross-entropy
d)
No special term
40.
Which GAN variant can generate MNIST digits of a specified class using label conditioning?
a)
DCGAN
b)
cGAN
c)
WGAN
d)
LSGAN
41.
CycleGAN is designed for:
a)
Text generation
b)
Unpaired image-to-image translation
c)
Stock prediction
d)
Classification
42.
CycleGAN introduces what key concept?
a)
Cycle consistency loss
b)
Binary encoding
c)
Dropout regularization
d)
Gradient clipping
43.
CycleGANs are best used when:
a)
Labeled paired data is available
b)
Only random noise is provided
c)
No paired data is available
d)
Time-series input only
44.
What does CycleGAN struggle with?
a)
Learning simple mappings
b)
Translating high-resolution images
c)
Mode collapse in limited data
d)
Data augmentation
45.
Which is a direct application of CycleGAN?
a)
Text-to-speech
b)
Style transfer between art genres
c)
Classifying news articles
d)
Data compression
46.
SGAN combines GANs with:
a)
Reinforcement learning
b)
Supervised classification task
c)
Clustering only
d)
Time-series analysis
47.
In SGAN, the discriminator outputs:
a)
Only binary output (real/fake)
b)
A multi-class output
c)
One-hot encoded regression
d)
Only labels
48.
SGAN is useful when:
a)
No labeled data
b)
Small labeled, large unlabeled data
c)
Only test data
d)
Data is sequential
49.
WGAN uses which loss to improve training?
a)
Least squares
b)
Wasserstein distance (Earth mover’s distance)
c)
Binary cross-entropy
d)
KL divergence
50.
A unique property of WGAN compared to vanilla GAN is:
a)
Uses fixed learning rate
b)
Uses weight clipping or gradient penalty
c)
Data normalization is mandatory
d)
Only for text data
100 %
