wayground logo

Free Printable Worksheets

Font size

S
M
L
XL
Worksheets

Generative AI Quiz

Total questions: 125

Worksheet time: 1hrs 3mins

Name
Class
Date
1.

How do researchers evaluate the performance of Generative AI models?

a)

By comparing generated outputs to predefined templates

b)

By conducting user surveys and feedback sessions

c)

By measuring similarity to real-world data distributions

d)

By applying subjective judgments and opinions

2.

What role does data play in the effectiveness of Generative AI models?

a)

No role, as Generative AI does not rely on data

b)

Limited role, with data serving as a minor input

c)

Significant role, with data influencing model performance

d)

Dominant role, with data determining model outcomes

3.

How might Generative AI contribute to addressing societal challenges?

a)

By exacerbating existing inequalities and injustices

b)

By promoting transparency and fairness in decision-making

c)

By fostering collaboration and community engagement

d)

By limiting access to generative technologies

4.

What ethical considerations are associated with the deployment of Generative AI?

a)

Bias, privacy, and accountability

b)

Transparency, fairness, and interpretability

c)

Inclusivity, diversity, and accessibility

d)

Innovation, creativity, and progress

5.

What impact might Generative AI have on traditional employment roles?

a)

No impact, as Generative AI is not suitable for real-world applications

b)

Negative impact, leading to job displacement in certain sectors

c)

Positive impact, creating new job opportunities across industries

d)

Limited impact, confined to specific job functions

6.

How does Generative AI contribute to innovation in various fields?

a)

By restricting access to creative tools

b)

By limiting experimentation and exploration

c)

By fostering creativity and generating novel ideas

d)

By promoting conformity and stagnation

7.

What role does creativity play in Generative AI?

a)

No role, as Generative AI relies solely on predefined rules

b)

Limited role, with creativity primarily driven by human input

c)

Significant role, with algorithms capable of creative output

d)

Minimal role, with creativity confined to specific tasks

8.

Which industries can benefit from the application of Generative AI techniques?

a)

Healthcare, finance, and education

b)

Manufacturing, agriculture, and transportation

c)

Entertainment, marketing, and fashion

d)

All of the above

9.

What advantages does Generative AI offer in comparison to traditional content creation methods?

a)

Faster production time and reduced costs

b)

Higher quality output with minimal human intervention

c)

Limited creativity and innovation

d)

Dependence on manual labor for each creation

10.

Which technique is used for generating new images based on artistic styles?

a)

Variational Autoencoders (VAEs)

b)

Deep Dream

c)

Neural Style Transfer

d)

Reinforcement Learning

11.

What distinguishes generative models from discriminative models?

a)

Generative models learn the underlying data distribution

b)

Discriminative models generate new data samples

c)

Generative models require labeled data for training

d)

Discriminative models predict probabilities

12.

Which of the following is NOT an application of Generative AI?

a)

Image classification

b)

Style transfer

c)

Music generation

d)

Text generation

13.

What is the main principle behind Generative Adversarial Networks (GANs)?

a)

Supervised learning

b)

Unsupervised learning

c)

Reinforcement learning

d)

Adversarial training

14.

What is a common application of generative chatbots?

a)

Image captioning

b)

Language translation

c)

Conversation generation

d)

Sentiment analysis

15.

Which type of neural network is commonly used for generating sequences of data?

a)

Convolutional Neural Networks (CNNs)

b)

Recurrent Neural Networks (RNNs)

c)

Generative Adversarial Networks (GANs)

d)

Reinforcement Learning Networks (RLNs)

16.

What is the primary focus of Generative AI?

a)

Classifying data

b)

Generating new data

c)

Predicting outcomes

d)

Optimizing algorithms

17.

In which domain is generative AI NOT commonly applied?

a)

Image synthesis

b)

Music generation

c)

Medical diagnosis

d)

Text generation

18.

Which type of neural network is commonly used in Generative AI?

a)

Convolutional Neural Networks (CNNs)

b)

Recurrent Neural Networks (RNNs)

c)

Generative Adversarial Networks (GANs)

d)

Reinforcement Learning Networks (RLNs)

19.

What distinguishes Variational Autoencoders (VAEs) from other generative models?

a)

They use reinforcement learning

b)

They are unsupervised learning models

c)

They learn a continuous latent space

d)

They are based on convolutional networks

20.

What does Generative AI focus on?

a)

Replicating existing data

b)

Creating new data

c)

Analyzing data patterns

d)

Optimizing algorithms

21.

What role does the generator play in a Generative Adversarial Network (GAN)?

a)

It discriminates between real and fake data

b)

It generates new data samples

c)

It optimizes the loss function

d)

It creates noise for training

22.

Which technique is used for transferring artistic styles onto images?

a)

Variational Autoencoders (VAEs)

b)

Deep Dream

c)

Neural Style Transfer

d)

Reinforcement Learning

23.

What distinguishes generative models from discriminative models?

a)

Generative models predict probabilities

b)

Discriminative models generate new data

c)

Generative models learn the joint probability distribution

d)

Discriminative models create realistic images

24.

Which of the following is NOT a common application of Generative AI?

a)

Image generation

b)

Music composition

c)

Speech recognition

d)

Text generation

25.

What is the main goal of Generative Adversarial Networks (GANs)?

a)

Image classification

b)

Data compression

c)

Generating realistic data

d)

Language translation

26.

What is the primary goal of an autoencoder?

a)

Classify input data into labels

b)

Compress and reconstruct input data

c)

Generate adversarial examples

d)

Predict time-series data

27.

In the context of latent space, which of the following is true?

a)

Latent space is the noise added during training

b)

It represents a compressed representation of input data

c)

It is the raw pixel space of images

d)

It always has a higher dimension than input data

28.

Which of the following is a common loss function used in training autoencoders?

a)

Cross-entropy loss

b)

Hinge loss

c)

Mean squared error (MSE)

d)

Huber loss

29.

What is the primary purpose of KL Divergence in Variational Autoencoders (VAEs)?

a)

To enforce data reconstruction

b)

To increase training speed

c)

To regularize the distribution of latent variables

d)

To increase model capacity

30.

What does the decoder do in an autoencoder architecture?

a)

Maps input data to latent space

b)

Reconstructs the input from latent space

c)

Computes the loss

d)

Adds noise to the input

31.

In PyTorch, which module is commonly used to define a custom neural network?

a)

torch.nn.Module

b)

torch.utils.data

c)

torch.TensorBoard

d)

torch.nn.functional

32.

Which of the following best describes the limitation of basic autoencoders in generative tasks?

a)

They can't compress data

b)

They do not learn a smooth latent space

c)

They overfit easily

d)

They use too much memory

33.

What does VAE sampling involve?

a)

Using deterministic latent variables

b)

Sampling from a learned distribution (mean and variance)

c)

Applying dropout to the encoder

d)

Selecting the maximum activation in each layer

34.

Which arithmetic operation is demonstrated in latent vector arithmetic?

a)

Convolution

b)

Addition and subtraction of vectors in latent space

c)

Multiplication of feature maps

d)

Max-pooling of features

35.

What is the role of the Denoising Autoencoder (DAE)?

a)

To generate new images

b)

To classify noisy input data

c)

To reconstruct clean data from noisy inputs

d)

To enforce sparsity in outputs

36.

In a VAE, the encoder outputs two vectors. What do they represent?

a)

Loss and accuracy

b)

Mean and log variance of the latent distribution

c)

Input and output labels

d)

Activation maps

37.

Which library in PyTorch is used for automatic differentiation?

a)

torch.optim

b)

torch.autograd

c)

torch.nn.functional

d)

torch.utils.data

38.

Which task is NOT typically handled by autoencoders?

a)

Denoising images

b)

Data compression

c)

Classification

d)

Feature extraction

39.

What is one application of VAEs in the medical field?

a)

Diagnosing infections

b)

Synthesizing medical images for augmentation

c)

Replacing doctors

d)

Performing surgeries

40.

Autoencoders can help reduce:

a)

Learning rate

b)

Memory usage

c)

Dimensionality of data

d)

Training epochs

41.

Autoencoders are often used in recommendation systems for:

a)

Label encoding

b)

User embedding generation

c)

Predicting weather

d)

Data normalization

42.

A use of autoencoders in NLP includes:

a)

Syntax correction

b)

Tokenization

c)

Sentence embedding

d)

Part-of-speech tagging

43.

Which property is desirable in a good latent space?

a)

Sparsity

b)

Semantic smoothness

c)

Noise amplification

d)

High dimensionality

44.

What happens when two points in latent space are interpolated?

a)

Noise is reduced

b)

Reconstruction is skipped

c)

A meaningful transition between representations is created

d)

Dropout is applied

45.

Latent space helps with:

a)

Overfitting

b)

Model compression

c)

Understanding data structure

d)

Enhancing activation functions

46.

Which model learns a latent distribution explicitly?

a)

CNN

b)

GAN

c)

VAE

d)

LSTM

47.

Which characteristic is true for continuous latent variables?

a)

They are binary

b)

They allow smooth interpolation

c)

They cannot be visualized

d)

They need labels

48.

Tied weights in an autoencoder mean:

a)

Encoder and decoder use different parameters

b)

Decoder reuses encoder weights

c)

Latent space is doubled

d)

Gradients are shared across epochs

49.

Which component is NOT essential in an autoencoder?

a)

Encoder

b)

Decoder

c)

Classifier

d)

Bottleneck

50.

Undercomplete autoencoders:

a)

Have fewer hidden units than input size

b)

Overfit the training data

c)

Do not use backpropagation

d)

Encode categorical variables only

51.

The purpose of sparsity constraints in autoencoders is:

a)

Increase accuracy

b)

Encourage feature selectivity

c)

Reduce training time

d)

Eliminate noise

52.

Overcomplete autoencoders:

a)

Compress input

b)

Use dropout aggressively

c)

Have more hidden units than input size

d)

Fail to generalize

53.

L1 loss promotes:

a)

Large errors

b)

Smooth outputs

c)

Sparsity in activations

d)

Overfitting

54.

Which loss is more sensitive to outliers?

a)

L2 loss

b)

Huber loss

c)

MSE

d)

Cross-entropy

55.

A good reconstruction loss should:

a)

Maximize overfitting

b)

Minimize the difference between input and output

c)

Add regularization noise

d)

Ignore small differences

56.

Custom loss functions in PyTorch require:

a)

torch.nn.Linear

b)

autograd.backward

c)

Subclassing nn.Module

d)

Defining forward logic with gradients

57.

In PyTorch, backward() is used to:

a)

Run forward pass

b)

Update weights manually

c)

Compute gradients

d)

Stop training

58.

Which PyTorch module is used to define custom neural networks?

a)

torch.nn.Module

b)

torch.nn.functional

c)

torch.optim

d)

torch.tensor

59.

What does model.train() do in PyTorch?

a)

Freezes weights

b)

Switches model to training mode

c)

Starts the optimizer

d)

Evaluates the model

60.

What is the purpose of torch.no_grad()?

a)

Enforces gradient computation

b)

Enables dropout

c)

Disables gradient calculation

d)

Trains the model

61.

What is the optimizer used for in PyTorch?

a)

Creating datasets

b)

Saving the model

c)

Updating model weights

d)

Building layers

62.

Which function is used to calculate loss in PyTorch?

a)

loss()

b)

compute_loss()

c)

criterion()

d)

nn.()

63.

One limitation of autoencoders is:

a)

They require supervised labels

b)

They cannot compress data

c)

They may memorize instead of generalizing

d)

They are only usable on text

64.

Why do autoencoders struggle with generative tasks?

a)

They lack latent space constraints

b)

They use too many layers

c)

They are too simple

d)

They rely on labels

65.

What issue occurs when the latent space is not regularized?

a)

Increased accuracy

b)

Clear class boundaries

c)

Poor generalization

d)

Faster training

66.

Autoencoders perform poorly when:

a)

The latent space is small

b)

The input data is low-dimensional

c)

They are trained with labels

d)

Reconstruction loss is high for training data

67.

Vanilla autoencoders lack:

a)

A decoder network

b)

A latent space

c)

A probabilistic framework

d)

Activation functions

68.

In a VAE, the encoder outputs:

a)

A reconstructed image

b)

One latent vector

c)

Mean and log-variance vectors

d)

Softmax probabilities

69.

The decoder in a VAE uses:

a)

Raw input data

b)

Random noise

c)

Samples from latent distribution

d)

Training labels

70.

Why do VAEs use the reparameterization trick?

a)

To increase randomness

b)

To compute gradients through stochastic sampling

c)

To speed up decoding

d)

To convert data to tensors

71.

What is a key advantage of VAEs over standard autoencoders?

a)

Higher compression

b)

Better reconstruction

c)

Structured latent space suitable for sampling

d)

Faster training time

72.

VAEs treat the latent space as:

a)

Discrete

b)

Randomized embeddings

c)

A probability distribution

d)

A loss function

73.

Which two distributions are compared using KL Divergence in VAEs?

a)

Input and output

b)

Prior and posterior latent distributions

c)

Noise and signal

d)

Labels and predictions

74.

A high KL Divergence value suggests:

a)

Latent distribution is similar to the prior

b)

Latent distribution diverges from the prior

c)

Model is regularized

d)

Loss is minimal

75.

KL Divergence is zero when:

a)

The decoder fails

b)

Prior equals posterior

c)

Input is noise

d)

Output is binary

76.

In VAE sampling, the latent vector is drawn from:

a)

A uniform distribution

b)

A one-hot vector

c)

A Gaussian distribution

d)

A categorical distribution

77.

The reparameterization trick enables:

a)

Manual weight updates

b)

Computation of gradients through sampling

c)

Saving models during training

d)

Using categorical data

78.

The random noise added during sampling in a VAE comes from:

a)

The decoder

b)

A learned distribution

c)

A standard normal distribution

d)

The output labels

79.

VAE sampling enables:

a)

Reconstruction without input

b)

Classification of data

c)

Generation of new samples

d)

Data shuffling

80.

Which operation is used in VAE sampling?

a)

z = mean + noise

b)

z = mean + std * epsilon

c)

z = decoder(input)

d)

z = encoder(output)

81.

Latent vector arithmetic involves:

a)

Normalizing pixel values

b)

Adding and subtracting embeddings

c)

Sorting datasets

d)

Random dropout

82.

In generative models, latent vector arithmetic can reveal:

a)

Training loss

b)

Hyperparameter tuning

c)

Semantic relationships

d)

Overfitting issues

83.

What kind of structure is expected in a good latent space?

a)

Linear separability

b)

Sparsity

c)

Discrete tokens

d)

Constant variance

84.

Latent vector arithmetic demonstrates that:

a)

Decoders ignore noise

b)

Latent space captures useful semantic features

c)

Models overfit easily

d)

Training data must be labeled

85.

An example of vector arithmetic in VAEs is:

a)

Adding noise to training labels

b)

Walking in the latent space to interpolate between concepts

c)

Removing layers from decoder

d)

Increasing KL divergence

86.

A Denoising Autoencoder is trained to:

a)

Add noise to data

b)

Learn a compressed vector

c)

Reconstruct the original input from noisy input

d)

Classify noisy inputs

87.

DAEs help models to:

a)

Memorize input features

b)

Overfit noisy labels

c)

Learn robust representations

d)

Ignore noise completely

88.

What kind of noise is commonly added in DAEs?

a)

Gaussian noise

b)

One-hot noise

c)

Binary noise

d)

Softmax noise

89.

The goal of DAEs is to:

a)

Increase dimensionality

b)

Remove dropout

c)

Generalize better by learning invariants to noise

d)

Increase activation sparsity

90.

DAEs differ from standard autoencoders by:

a)

Using no encoder

b)

Training with noisy inputs

c)

Removing loss functions

d)

Only working on text data

91.

Which of the following fields uses autoencoders for anomaly detection?

a)

E-commerce

b)

Network security

c)

Weather prediction

d)

Game design

92.

Autoencoders are useful in image colorization because they:

a)

Reduce color space

b)

Learn grayscale compression

c)

Reconstruct missing chromatic channels

d)

Encode RGB values

93.

In fraud detection, autoencoders help by:

a)

Predicting transaction amounts

b)

Labeling fraud directly

c)

Detecting outliers in encoded space

d)

Encrypting user data

94.

Which application uses VAE for data augmentation?

a)

Text summarization

b)

Image synthesis

c)

Sorting algorithms

d)

Language translation

95.

Autoencoders can be applied in sound processing to:

a)

Track speaker age

b)

Denoise audio signals

c)

Translate music to text

d)

Predict lyrics

96.

Which loss function is common for binary classification?

a)

MSE

b)

Categorical cross-entropy

c)

Binary cross-entropy

d)

L1 loss

97.

Reconstruction loss in an autoencoder should:

a)

Be high for regular inputs

b)

Minimize the difference between input and output

c)

Maximize the KL divergence

d)

Penalize sparsity

98.

Which loss function is most commonly used with VAEs?

a)

MSE only

b)

BCE + KL Divergence

c)

L2 norm only

d)

L1 norm + Dropout loss

99.

Why is KL Divergence added to the loss in VAEs?

a)

To penalize random noise

b)

To regularize the latent space

c)

To increase output sharpness

d)

To avoid vanishing gradients

100.

Huber loss is ideal when:

a)

You want to ignore all outliers

b)

You expect binary classification

c)

You need a compromise between L1 and L2 losses

d)

Your output is probabilistic

101.

In PyTorch, which function performs a forward pass?

a)

.loss()

b)

.step()

c)

.forward()

d)

.train()

102.

How are gradients zeroed in PyTorch?

a)

optimizer.zero_grad()

b)

model.zero_grad()

c)

loss.backward()

d)

torch.reset()

103.

Which module provides activation functions like ReLU in PyTorch?

a)

torch.nn.functional

b)

torch.optim

c)

torch.cuda

d)

torch.init

104.

What does model.eval() do in PyTorch?

a)

Trains the model

b)

Resets weights

c)

Turns off dropout and batchnorm

d)

Compiles the model

105.

What is required to perform backpropagation in PyTorch?

a)

.compile()

b)

.eval()

c)

.zero_grad()

d)

.backward()

106.

Which problem is common when autoencoders are too powerful?

a)

Underfitting

b)

Memorizing training data

c)

Reducing input noise

d)

Over-regularization

107.

Autoencoders are not ideal for:

a)

Dimensionality reduction

b)

Data generation without structure

c)

Anomaly detection

d)

Denoising

108.

What is a key drawback of deterministic autoencoders?

a)

They are slow to train

b)

They cannot reconstruct images

c)

They lack stochastic sampling ability

d)

They do not use loss functions

109.

Why might an autoencoder fail in generalization?

a)

Excessive KL divergence

b)

Too small batch size

c)

Lack of latent space constraints

d)

Use of ReLU activation

110.

A potential issue with autoencoders in practice is:

a)

Latent space being interpretable

b)

Over-regularization from KL

c)

Poor interpolation between samples

d)

Limited hardware support

111.

What does the encoder in a VAE learn?

a)

Deterministic latent code

b)

A probability distribution over latent variables

c)

A one-hot label

d)

The decoder function

112.

What does KL divergence prevent in a VAE?

a)

Overfitting by decoder

b)

Irregularity in latent space

c)

Sampling variability

d)

Softmax overfitting

113.

The decoder in a VAE learns to:

a)

Reconstruct the noise

b)

Match samples to true labels

c)

Map latent samples to data space

d)

Compress data into a bottleneck

114.

What improves sampling in VAEs?

a)

Randomly shuffling weights

b)

Encouraging the latent distribution to match a known prior

c)

Adding dropout to latent layer

d)

Removing reconstruction loss

115.

A sign that VAE training is failing is:

a)

Latent vectors are Gaussian

b)

KL loss is near zero constantly

c)

Encoder uses two outputs

d)

Reconstruction loss decreases

116.

KL Divergence is asymmetric, meaning:

a)

D(P‖Q) = D(Q‖P)

b)

D(P‖Q) ≠ D(Q‖P)

c)

It does not measure distributions

d)

It is always zero

117.

KL Divergence is minimized when:

a)

P(x) = 0 for all x

b)

P and Q are independent

c)

P and Q are identical

d)

Q has lower entropy

118.

In VAE training, KL divergence penalizes:

a)

Output variance

b)

Overfitting to the decoder

c)

Deviation from prior distribution

d)

Noise in the dataset

119.

KL Divergence measures the information lost when:

a)

Using one distribution to approximate another

b)

Compressing data

c)

Encoding inputs as one-hot

d)

Discarding outliers

120.

A key mathematical property of KL Divergence is:

a)

It's bounded above by 1

b)

It's negative when distributions overlap

c)

It's always non-negative

d)

It works only for discrete data

121.

Latent space refers to:

a)

A visible layer in a neural network

b)

Intermediate compressed representation of input

c)

Raw input data

d)

Fully connected output layer

122.

In a VAE, points in latent space correspond to:

a)

Random noise

b)

One-hot vectors

c)

Encoded representations of data

d)

Class labels

123.

A good latent space should:

a)

Be unstructured

b)

Contain only zero vectors

c)

Allow smooth interpolation between data points

d)

Be trained using supervised learning only

124.

Which of the following helps regularize latent space?

a)

Reconstruction loss

b)

KL Divergence

c)

Dropout

d)

Batch normalization

125.

Latent space enables:

a)

Labeling unstructured data

b)

Visualizing class probabilities

c)

Generative capabilities in VAEs

d)

Overfitting reduction in CNNs