GEN AI

Introduction to Generative AI - Prequiz

1. Which statement best defines Generative AI?
AI that can generate new data samples
AI that automates repetitive tasks
AI that understands human emotions
AI that predicts future trends
AI that classifies data

2 .What does AI stand for?

Application Interface
Automated Information
Automated Interaction
Artificial Intelligence
Advanced Integration

3. Which of the following is a real-world example of GenerativeAI?

Generating realistic human faces in movies
Translating languages
Predicting stock market prices
Automating cars
Sorting emails

4. In which application is Generative AI NOT typically used?

Creating art
Producing realistic video game characters
Designing virtual environments
 Generating music
Automating customer service chats

5. Which of the following is NOT a type of AI?

Generative AI
Unsupervised AI
Supervised AI
Generative Art
Reinforcement AI

6. Generative AI is closely related to which type of models?

Clustering models
Classification models
Decision trees
Regression models
Generative models

7. Which AI type primarily focuses on labeling data?

Semi-supervised AI
Supervised AI
Reinforcement AI
Regression AI
Generative AI

8. Which of the following fields can utilize Generative AI to create

new, original content or simulations?
Banking
Transportation
 Data Analysis
E-commerce
Art and Music

9. What distinguishes Generative AI from Discriminative AI?

Generative focuses on labeling, Discriminative on generating
Generative is older, Discriminative is newer
Both are the same
Generative is for images, Discriminative for text
Generative models data distribution, while Discriminative
models the boundary between classes

10. Which type of AI is primarily concerned with how data is

generated rather than how it's separated?
Supervised Learning
Reinforcement Learning
Discriminative AI
Generative AI
Unsupervised Learning

Introduction to Generative AI - Post quiz

1. Why is Generative AI considered significant in the realm of
artificial intelligence?
It simplifies complex algorithms
It reduces the need for large datasets
It exclusively works with images
It can produce new, previously unseen data samples
It speeds up training processes
2. Which of the following is a direct application of Generative AI in
the entertainment industry?
Creating realistic CGI characters
Translating movie scripts
Automating video editing
Recommending movies to users
Predicting movie success

3. What is Generative AI primarily used for?

Data labeling
Optimization
Generating new data
Regression
Classification

4. Which is NOT a real-world application of Generative AI?

Deep fake videos
Producing synthetic voices
Creating virtual fashion designs
Predicting stock market prices
Generating game environments

5. How does Generative AI differ from Classification AI?

It's faster
It generates new data rather than categorizing existingdata
It requires more data
It's more accurate
It's easier to implement
6. If an AI system is designed to label images of cats and dogs, it
is primarily a _______ model.
Unsupervised
Discriminative
Reinforcement
Hybrid
Generative

7. In the context of AI, which model type is more concerned with

the underlying distribution of data?
Classification AI
Hybrid AI
Generative AI
Regression AI
Reinforcement AI

8. Generative AI can be used to create which of the following?

Regression models
Decision boundaries
Data labels
Classification categories
New artworks and music pieces

9. Which statement best describes the role of Generative AI?

It is primarily used for data sorting
It is exclusively used in robotics
It is best suited for regression tasks
 It is the oldest form of AI
It focuses on generating data based on learned patterns

10. Which AI type is best for predicting outcomes?

Regression AI
Classification AI
Semi-supervised AI
Reinforcement AI
Generative AI

Brief History of Generative AI - Pre Quiz

1. In which year were Generative Adversarial Networks (GANs)
first introduced?
2012
2018
2016
2010
2014

2. What is the primary purpose of generative models?

Generating new data
Classifying data
Recognizing patterns
Filtering data
None of the given options
3. Which architecture is primarily associated with attention
mechanisms?
CNN
Transformer
RNN
VAE
GAN

4. Which model uses a probabilistic approach to encode and

decode data?
BigGAN
DCGAN
CycleGAN
Transformer
VAE

5. Which model marked a significant milestone in the use of

transformers in NLP?
BERT
GAN
LSTM
CNN
RNN

6. Which of the following is NOT a direct application of GANs but

rather an outcome of its influence?
Image-to-Image translation
Super-resolution
 Style transfer
Generating realistic images
Reinforcement learning in game playing

7. Which pioneering research in Generative AI specifically

emphasized the generation of text sequences?
"Sequence to Sequence Learning with Neural Networks"
"Understanding Machine Learning: From Theory to Algorithms"
"Visualizing and Understanding Convolutional Networks"
"A Neural Algorithm of Artistic Style"
"DeepFace: Closing the Gap to Human-Level Performance in
Face Recognition"

8. Who introduced Generative Adversarial Networks (GANs)?

Andrew Ng
Yoshua Bengio
Geoffrey Hinton
Ian Goodfellow
Yann LeCun

9. What are the two main components of a GAN?

None of the given options
Forward and Backward
Encoder and Decoder
Input and Output
Generator and Discriminator
10. Which of the following research papers is foundational for
Variational Autoencoders (VAEs)?
"Deep Residual Learning for Image Recognition"
"Generative Adversarial Nets"
"Auto-Encoding Variational Bayes"
"Mastering Chess and Shogi by Self-Play"
"Attention is All You Need"

Brief History of Generative AI - Post Quiz

1. What is the primary advantage of Transformers over RNNs in
terms of processing sequences?

Better attention mechanism

None of the given options
More parameters
Parallel Processing
Faster convergence

2. Which AI model series by OpenAI, based on the Transformer

architecture, is known for generating highly coherent content?

ResNet
BERT
GPT series
CycleGAN
TransformerXL

3. In the context of GANs, what is the role of the Discriminator?

To transform data
 To generate data
To distinguish between real and generated data
To encode data
To decode data

4. Which model demonstrated that using larger architectures can

produce better images?

BigGAN
CycleGAN
DCGAN
VAE
Transformer

5. Which model can transform horse photos into zebra photos

without direct comparison?

DCGAN
Transformer
CycleGAN
VAE
BigGAN

6. What mechanism allows the Transformer model to weigh the

importance of different words in a sequence?

Decoding Mechanism
Self-Attention Mechanism
None of the given options
Encoding Mechanism
Recurrent Mechanism
7. What is the main innovation introduced by the "Attention Is All
You Need" paper?

Introduction of RNNs
Introduction of CNNs
Introduction of GANs
Transformer architecture
Introduction of VAEs

8. Which generative model introduced a stochastic layer that

models data in a latent space?

Transformer
DCGAN
BigGAN
CycleGAN
VAE

9. Which of the following is NOT a direct application of the

Transformer architecture?
Text translation
Question answering
Image recognition
Text summarization
Image generation

10. Which model is known for its rules for creating stable and
effective AI image-makers?
 CycleGAN
VAE
DCGAN
BigGAN
Transformer

Fundamentals of ML - Pre QuiZ

1. Which of the following is NOT a common machine learning
algorithm?

K-Means Clustering
Support Vector Machines
Decision Trees
Neural Networks
Quantum Entanglement

2. Which of the following is a technique to prevent overfitting in

neural networks?

Using a larger dataset

Learning Rate Adjustment
Dropout
Increasing the number of layers
Gradient Clipping

3. What is the primary purpose of a loss function in training neural

networks?

To define the network's architecture

To speed up training
 To quantify the difference between predicted and actual
values
To initialize weights
To activate neurons

4. Which activation function outputs a value between 0 and 1?

Leaky ReLU
Sigmoid
Hyperbolic Tangent (tanh)
Linear
Rectified Linear Unit (ReLU)

5. In the context of neural networks, what does the term

"backpropagation" refer to?

The initial random assignment of weights

The method of adjusting weights based on the error
The forward flow of data
The activation of neurons in the hidden layer
The process of adding more layers

6. Which component of a neural network is responsible for

combining inputs and passing them to the next layer?

Neuron (or Node)

Layer
Weight
Activation Function
Bias
7. What is the main difference between regression and
classification?

Regression predicts a continuous output, Classification

predicts a discrete label
Classification is unsupervised
Regression uses labeled data, Classification doesn't
Regression is unsupervised
Both are the same

8. Which of the following is NOT a type of machine learning?

Supervised Learning
Recursive Learning
Unsupervised Learning
Reinforcement Learning
Semi-supervised Learning

9. What is the primary goal of machine learning?

To program explicit rules for a task

None of the given options
To increase computational speed
To design new algorithms
To allow computers to learn from data

10. Which application of ML is used to group similar items?

Recommendation
Ranking
Regression
 Classification
Clustering

Fundamentals of ML - Post Quiz

1. Which function introduces non-linearity in a neural network?

Bias Function
Loss Function
Activation Function
Weight Function
Linear Function

2. In a neural network, what does a neuron compute?

The gradient of the loss

The error of the network
A weighted sum followed by an activation function
A fixed value
The learning rate

3. How is a neural network's performance typically evaluated

during training?

Using the training data

Using a validation set
Using the weights
Using the test data
Using the activation functions
4. Which of the following is a challenge in training deep neural
networks?

Too few neurons

Vanishing/Exploding gradients
Linear activation functions
Small datasets
All neurons activating at once

5. What is the primary purpose of backpropagation?

Forward propagation of data

Activation of neurons
Data preprocessing
Adjusting weights based on the error
Initialization of weights

6. What is the role of the loss function in training a neural

network?

To activate the neurons

To quantify the difference between predicted and actual
values
To define the network architecture
To introduce non-linearity
To initialize the weights

7. Which of the following is NOT a layer type in a typical neural

network?

Hidden Layer
 Quantum Layer
Output Layer
Convolutional Layer
Input Layer

8. In which type of ML does an agent learn by interacting with an

environment?

Clustering
Supervised Learning
Unsupervised Learning
Regression
Reinforcement Learning

9. Which of the following is a common activation function in neural

networks?

Polynomial Function
ReLU (Rectified Linear Unit)
Linear Function
Bias Activation
Weighted Sum

10. Which application of ML is used to detect unusual patterns in

data?

Regression
Ranking
Clustering
Classification
Anomaly Detection
Quiz: Introduction to Generative Models - Pre Quiz
1. Which statement best differentiates generative from
discriminative models?

Generative models are newer than discriminative models

Generative models are only for images, discriminative for text
Both models serve the same purpose
Generative models cannot be trained with labeled data
Generative models learn the joint probability distribution,
while discriminative models learn the conditional probability

2. In the context of generative models, what does P(x) represent?

The probability distribution of the data x

The posterior probability of x
The conditional probability of x given y
The likelihood of x
The joint probability of x and y

3. Which of the following is NOT a generative model?

Restricted Boltzmann Machines

Generative Adversarial Networks
Gaussian Mixture Models
Support Vector Machines
Variational Autoencoders

4. What does likelihood measure in the context of a model?

 The complexity of the model
The error rate of the model
The generative capacity of the model
How well the model explains the observed data
The probability of the model being correct

5. If a model is better at distinguishing between classes rather

than generating data, it is likely a _______.

Generative model
Bayesian model
Likelihood model
Joint probability model
Discriminative model

6. Which model type is primarily concerned with determining P(y |

x)?

Discriminative Model
Both Generative and Discriminative
Bayesian model
Generative Model
Probability Distribution

7. What is the primary goal of generative models in AI?

To optimize algorithms
To classify data
To generate new data samples
To analyze data distributions
To reduce computational cost
8. In the context of models, what does P(x | y) typically represent?

The probability of x given y

The distribution of y
The probability of y given x
The generative capacity of x
The likelihood of y

9. Generative models are primarily used for which of the following

tasks?

Regression
Classification
Generating new data samples similar to the input data
Reinforcement learning
Clustering

10. Which of the following is crucial for understanding the

behavior of generative models?

Convolutional layers
Activation functions
Probability distributions and likelihood
Gradient descent
Backpropagation
Introduction to Generative Models - Post Qiuz
1. Which of the following is NOT a property of likelihood?

It is a function of model parameters

It can be used to compare different models
It measures how well a model explains data
It is not normalized like a probability
It is always a probability between 0 and 1

2. Which of the following best describes the difference between

generative and discriminative models?

Discriminative models can't generate data

Generative models are always better
Generative models are used for classification only
Generative models learn the data distribution, while
discriminative models learn the decision boundary
Generative models are older in concept

3. Which claim regarding generative models isn't true?

They can generate new data samples

They always require labeled data for training
They capture the data distribution
They can be combined with discriminative models for certain
tasks
They can be used in unsupervised learning scenarios

4. Which model type aims to capture the joint probability P(x, y)?

Discriminative Model
 Both Generative Model and Discriminative Model
Generative Model
Probability Distribution
regression model

5. How is the likelihood of data given a model symbolized?

P(model | data)
P(data)
P(data & model)
P(model)
P(data | model)

6. What does a probability distribution provide?

A measure of model error

A decision boundary for classification
A method for generating new data
A mathematical description of outcomes for a random
variable
A training method for models

7. What's a significant hurdle when training GANs?

The discriminator becoming too weak

Inability to generate high-resolution images
Overfitting to the training data
Slow convergence rate
Mode collapse
8. Within the architecture of Generative Adversarial Networks
(GANs), which duo of fundamental elements are paramount?

Classifier and Regressor

Forward and Backward Propagators
Encoder and Decoder
Generator and Discriminator
Activator and Deactivator

9. Within generative models, what function does the discriminator

serve in GANs?

To distinguish between real and generated data

To capture the joint probability
To calculate the likelihood
To optimize the generator
To generate new data

10. For what tasks can generative models be applied?

Only denoising
Only data generation
Data generation, denoising, inpainting, and more
Classification only
Data labeling only

Variational Autoencoders - Pre Quiz

1. Reparameterization trick is used to...

Improve model accuracy

 Deal with the non-differentiability of sampling in VAEs
Reduce model size
None of the given options
Speed up training

2. Which component of the VAE loss function ensures the latent

variables adhere to a standard distribution?

Hinge loss
Absolute error
Cross-entropy
KL divergence
Mean squared error

3. Why are autoencoders considered generative models?

They are used for supervised learning

They are only used for image data
They can reconstruct and generate data similar to the
input
They always reduce data dimensionality
They are a type of neural network

4. What is the primary role of autoencoders in generative

modeling?

Data compression and reconstruction

Clustering
Data classification
Image recognition
Regression
5. Why is the reparameterization trick crucial in training VAEs?

It increases the model's accuracy

It speeds up the training process
It allows backpropagation through stochastic nodes
It reduces the need for labeled data
It reduces the model's complexity

6. What does VAE stand for?

None of the given options

Vectorized Autoencoder
Virtual Autoencoder
Variational Autoencoder
Variable Autoencoder

7. Which of the following is NOT a type of autoencoder?

Sparse autoencoder
Supervised autoencoder
Variational autoencoder
Denoising autoencoder
Contractive autoencoder

8. In the context of Variational Autoencoders (VAEs), what does

variational inference help achieve?

Direct computation of posterior distributions

Reduction of model parameters
 Faster training speeds
Improved image resolution
Approximation of complex posterior distributions

9. In which application might you use a VAE for generating new,

coherent samples?

Speech recognition
Designing virtual fashion items
Text translation
Time series forecasting
Image classification

10. Which application does NOT typically use VAEs?

Face generation for video games

Fashion design
Medical imaging enhancement
Text summarization
Anomaly detection in industrial equipment

Quiz: Variational Autoencoders - Post Quiz

1. In which application might VAEs be used to enhance image
quality?
Video streaming
Social media photo filters
Medical imaging
None of the given options
Text generation
2. Why is variational inference used in VAEs?

To approximate intractable posterior distributions

To speed up training
None of the given options
To reduce model size
To improve model accuracy

3. Which of the following is a key component of the VAE loss

function?

Accuracy
Cross-entropy only
Precision
KL divergence
Mean squared error only

4. In which application can VAEs detect unusual patterns?

Text generation
Face generation for video games
Fashion design
Music composition
Anomaly detection in industrial equipment

5. What do VAEs use to generate a distribution over latent

variables?

Backpropagation
Variational inference
 Gradient descent
Transfer learning
None of the given options

6. Which optimization technique is commonly used with VAEs?

Genetic algorithms
Simulated annealing
Principal component analysis
Stochastic gradient descent (SGD)
None of the given options

7. Which of the following is NOT a typical use case for VAEs?

Real-time speech translation

Medical imaging enhancement
Anomaly detection in industrial equipment
Face generation for video games
Fashion design

8. Why is the reparameterization trick important in VAEs?

It increases model efficiency

It simplifies the model architecture
It allows backpropagation through random nodes
It reduces overfitting
None of the given options

9. Autoencoders primarily focus on which aspect of data?

Generation
 Classification
Reconstruction
Clustering
Filtering

10. How do VAEs differ from traditional autoencoders?

VAEs use supervised learning

VAEs are simpler
VAEs introduce randomness via a probabilistic layer
VAEs are faster
VAEs are more accurate

CASE STUDY - VAE Application – Quiz

1. What is the y-axis label of the chart visualizing the error?

Latent Space
Anomaly Score
Timestamp
Data Value
Reconstruction Error

2. Why is data preprocessing required before training the VAE?

To make the data look visually appealing

To introduce errors into the data
To make the data unreadable
To ensure it is suitable for training
To make the data larger
3. How is the data divided for training the VAE?

80-20
90-10
70-30
60-40
50-50

4. What does the VAE attempt to minimize during training?

Validation accuracy
Latent space dimensions
Data input size
Loss
Training time

5. What criterion is used to determine if a data point is

anomalous?

If its error is above median error

If its error is in the top 10%
If its error is above the 99th percentile
If its error is below mean error
If its error is above mean error

6. Why is understanding the VAE's outputs challenging?

They are too simplistic

They can be complex and non-intuitive
They are highly interpretable
They use an unknown language
 They are always correct

7. Which is NOT a challenge in implementing VAEs for this use-

case?

Increasing data storage costs

Data Quality
Latency
Model Training
Threshold Setting

8. For how many epochs is the VAE trained?

100
50
10
25
40

9. What is the VAE trained to learn effectively?

A compressed representation of the data

A visual representation of the data
A highly detailed representation of the data
A noisy representation of the data
A textual description of the data

10. In the VAE, what does the sampling function introduce?

Randomness
Recursion
 Parallelism
Determinism
Linearity

11. Over time, due to certain changes, what might be required of

the VAE model?

Disintegration
Reformatting
Shrinking
Continuous adaptation
Manual recalibration

12. What two components combine to form the VAE's loss?

MSE and KL divergence

MSE and Cross-entropy
L1 loss and L2 loss
KL divergence and Cross-entropy
Classification error and Regression loss

13. Which of the following is NOT an attribute in the given data?

Humidity
Pressure
Timestamp
Temperature
Vibration
14. What is a primary application of VAEs mentioned in the case
study?

Speech Recognition
Object Detection
Text Summarization
Image Classification
Anomaly Detection

15. What type of dataset does the manufacturing plant collect?

Text Dataset
Image Dataset
Tabular Dataset
Audio Dataset
Time Series Dataset

Generative Adversarial Networks - Pre Quiz

1. The training process of GANs is often likened to which game?

Chess
Sudoku
None of the given options
Poker
Minimax

2. In GANs, which component is responsible for evaluating the

authenticity of data?

Discriminator
 Encoder
Generator
None of the given options
Decoder

3. Progressive GANs are designed to address which challenge in

traditional GANs?

Inability to generate colored images

Training stability and generating high-resolution images
Slow training speeds
Discriminator overpowering the generator
Mode collapse

4. Which of the following is a real-world application where GANs

have shown significant promise?

Speech recognition
Image-to-image translation
Text summarization
Time series forecasting
Image classification

5. In the GAN architecture, what is the primary goal of the

Discriminator?

Generate realistic data samples

Minimize the loss function
Replicate the generator's output
Ensure mode diversity
Distinguish between real and generated samples
6. Which type of GAN allows for generating data based on
specific categories?

Progressive GAN
Conditional GAN
None of the given options
Mode Collapse GAN
Minimax GAN

7. What does GAN stand for?

Gradient Augmented Network

Generalized Artificial Network
Generative Adversarial Network
None of the given options
Generative Analytical Network

8. What is mode collapse in the context of GANs?

When the generator produces limited varieties of outputs

When the discriminator becomes too powerful
When the model converges too quickly
When the model overfits
When the model underfits

9. What is a challenge faced during GAN training due to the

minimax game concept?

Generator producing only a single mode

Overfitting to the training data
 Discriminator becoming too weak
Oscillations and non-convergence
Quick convergence to a suboptimal solution

10. Which component of a GAN is responsible for generating new

data samples?

Optimizer
Generator
Decoder
Encoder
Discriminator

Generative Adversarial Networks - Post Quiz

1. Which statement about GANs is true?

They can generate new, previously unseen data

None of the given options
They always converge to a solution
They only work with images
They are a type of supervised learning

2. The generator's objective in GANs is to...

Reduce mode collapse

None of the given options
Classify real vs. fake
Fool the discriminator
Improve model accuracy
3. In GANs, if the discriminator becomes too powerful, what can
happen?

The model achieves perfect accuracy

The generator becomes powerful too
The generator may struggle to improve
None of the given options
The training process speeds up

4. Which component of a GAN tries to produce fake data?

Decoder
Discriminator
Encoder
None of the given options
Generator

5. What is a challenge in evaluating the performance of GANs?

They always outperform other models
They are too fast
They require large datasets
Determining the quality of generated data
None of the given options

6. In the minimax game of GANs, what is the discriminator's goal?

Minimize its own loss

Distinguish between real and fake data
None of the given options
Maximize the generator's loss
 Generate realistic data

7. Which is NOT a real-world application of GANs?

Data augmentation

Real-time weather prediction

Super-resolution imaging
Style transfer
Art generation

8. Mode collapse is problematic because...

It requires more data

It makes the discriminator weak
None of the given options
It limits the diversity of generated outputs
It speeds up training

9. Which GAN variant can be conditioned on labels to generate

specific outputs?

Minimax GAN
Mode Collapse GAN
Conditional GAN
None of the given options
Progressive GAN

10. Which GAN variant focuses on gradually increasing the

resolution of generated images?
 None of the given options
Minimax GAN
Conditional GAN
Progressive GAN
Mode Collapse GAN

CASE STUDY - GANS - CIFAR – Quiz

1. Which activation function is used in the final layer of the
generator model?

tanh
relu
sigmoid
softmax
leakyrelu

2. In the discriminator's code, which layer helps in reducing the

dimensions of the input image?

Dense
Reshape
Conv2D with strides
BatchNormalization
UpSampling2D

3. What are the two main components of a GAN?

Discriminator & Sampler

Generator & Evaluator
Generator & UpSampler
 Discriminator & Evaluator
Generator & Discriminator

4. How many images are there in each class of the CIFAR-10

dataset?

10000
5000
12000
6000
15000

5. Which technique can help in dealing with training instability in

GANs?

Dropout
All of the given options
Data augmentation
Gradient clipping
Noise addition

6. Which challenge refers to the generator producing limited

varieties or even the same sample every time?

Mode Collapse
All of the given options
Data Augmentation
Convergence Issues
Training Instability
7. In the provided code, why is discriminator.trainable set to False
when setting up the combined system?

To speed up training
To prevent overfitting
None of the given options
To make sure only the generator is trained in this step
To increase discriminator's accuracy

8. In the generator code, what is the purpose of the Reshape

layer?

To critique the images

To normalize the image values
To flatten the images
To reshape the dense layer into a 3D tensor for images
To upsample the images

9. What does the discriminator do in a GAN?

Enhances image resolution

Combines images
Both create and evaluate images
Creates images
Evaluates if an image is real or fake

10. Which of the following best describes the role of the generator
in a GAN?

None of the given options

To combine images
 To produce images
To critique images
To evaluate the loss

11. Which architecture can help address convergence issues in

traditional GANs?

LSTM
WGAN
DBN
CNN
RNN

12. Why might someone want to use GANs on the CIFAR-10

dataset?

To classify the images in the dataset

To delete images from the dataset
To generate novel and relevant images to augment dataset
To critique the images in the dataset
To reduce the size of the dataset

13. What is used to refine the models during training?

Adam Optimizer
Conv2D
All of the given options
Batch Normalization
LeakyReLU
14. Which of the following is NOT a feedback given to the
generator during training?

This image looks blurry

This image is pixelated
This is a fake image
This is a genuine image
This image looks like a car

15. During training, what does the generator use to improve itself?

CIFAR-10 dataset
Feedback from the discriminator
Real images
Feedback from the user
Feedback from both the user and the discriminator

Sequence Generation with RNNs - Pre Quiz

1. Which of the following is NOT a type of RNN architecture?

GRU
LSTM
Bidirectional RNN
CNN
Simple RNN

2. In NLP, what does RNNs help to predict?

Next word
None of the options given
Next song note
 Next image
Next video frame

3. Which problem in RNNs does LSTM help to address?

All of the options given

High variance
Overfitting
Bias
Vanishing gradient

4. RNNs are primarily used for which type of data?

Audio
Tabular
None of the options given
Sequential
Image

5. What is the key advantage of using LSTMs over basic RNNs in

sequence generation tasks?

Lower computational cost

Faster training speeds
Less prone to overfitting
Ability to remember long-term dependencies
Simpler architecture

6. Which RNN architecture utilizes update and reset gates to

manage memory?
 LSTM
Echo State Network
Bidirectional RNN
Hopfield Network
GRU

7. When using RNNs for music generation, what does each

neuron in the output layer typically represent?

A frequency band
A specific instrument
A note in the C major scale
A time step in the generated sequence
A possible note or rest in the musical vocabulary

8. In the context of natural language processing, how are RNNs

typically utilized for machine translation?

As discriminators in GANs
Encoding the input sequence and decoding the output
sequence
For clustering text data
As a replacement for CNNs
For image classification

9. During the training of RNNs for sequence generation, what is

the common technique used to mitigate the vanishing gradient
problem?

L1 regularization
Dropout
 Data augmentation
Batch normalization
Gradient clipping

10. What does RNN stand for?

None of the options given

Random Neural Network
Recurrent Neural Network
Regular Neural Network
Recursive Neural Network

Sequence Generation with RNNs - Post Quiz

1. Which of the following is a common application of RNNs in
NLP?

Image generation
Image classification
Object detection
Text generation
Face recognition

2. Which problem arises when training RNNs on long sequences?

All of the given options

Vanishing or exploding gradients
High bias
Overfitting
Underfitting
3. In sequence generation tasks, what is the primary input to an
RNN at each time step?

Previous output
Current input
None of the given options
Previous error
Current weight

4. What is the main advantage of LSTM over basic RNN?

More layers
None of the given options
Handling long-term dependencies
Lower computational cost
Faster computation

5. How do RNNs handle variable-length sequences in NLP?

They don't
None of the given options
Through padding and truncation
By skipping them
By changing the network size

6. Which of the following is NOT a typical use case for RNNs?

Speech recognition
None of the given options
Text generation
 Image classification
Time series prediction

7. What is the primary difference between LSTM and GRU?

LSTM has 3 gates, GRU has 2

LSTM is older, GRU is newer
LSTM is faster, GRU is slower
LSTM is for sequences, GRU is for images
LSTM has input, forget, and output gates; GRU has reset
and update gates

8. Why might one use GRU over LSTM?

LSTM is outdated
LSTM can't handle sequences
None of the given options
GRU is simpler and sometimes faster
GRU is always more accurate

9. Which RNN architecture uses a reset and update gate?

GRU
Bidirectional RNN
None of the given options
Simple RNN
LSTM

10. In music generation, what might an RNN be trained to

predict?
 Next note or chord
None of the given options
Next song genre
Next album cover
Next instrument

Sentiment Analysis with RNNs - Case study

1. What is the role of the `<OOV>` token?

Delete out-of-vocabulary words

Regular expression matcher
Placeholder for out-of-vocabulary words
Placeholder for numbers
Ignore out-of-vocabulary words

2. Which layer in the RNN model represents words as detailed feature lists?

LSTM Layer
Dense Layer
SimpleRNN Layer
Dropout Layer
Embedding Layer

3. Why is padding used in the preprocessing step?

To handle variable review length

To reduce memory usage
To increase vocabulary size
For beautification
To improve accuracy
4. What advantage does LSTM have over traditional RNNs?

Simpler architecture
Lower memory usage
Requires fewer layers
Tackles the vanishing gradient problem
Faster convergence

5. What is the purpose of the Dropout layer in the LSTM with Dropout model?

Regularization to prevent overfitting

Tokenization
Recurrence
Activation function
Embedding

6. What might be a concern if the training accuracy is high but validation

accuracy is significantly low?

Model is perfectly trained

Model is underfitting
Data is incorrectly labeled
Model is overfitting
Model needs more layers

7. In which scenario might you prefer a simple RNN over an LSTM?

When high accuracy is a must

Long-range dependencies in data
Fast training with limited resources
Large datasets
Complex sentence structures
8. Which parameter in `model.fit()` signifies the number of times the model is
exposed to the dataset?

epochs
loss
optimizer
validation_data
batch_size

9. Why is the loss function important during model compilation?

Specifies number of epochs

Assigns weights to layers
Specifies how errors are measured
Determines model layers
Adjusts learning rate

10. How does the model handle reviews of varying lengths?

Changes tokenizer's vocabulary

Ignores reviews outside a certain length range
Uses LSTM layers
Uses multiple RNN layers
Uses padding

11. Why might the vanishing gradient problem be a challenge in RNNs?

Increases accuracy
Reduces training speed
Impedes learning of long-range dependencies
Requires more memory
 Makes model evaluation faster

12. In the given LSTM model, which layer(s) help in retaining memory and
context?

Dropout layer
Embedding layer
SimpleRNN layer
LSTM layer
Dense layer

13. When using a tokenizer with a fixed number of words, what could be a potential
drawback?

Simplifies the model

Enhances accuracy
Slows down training
Increases memory usage
Limited understanding due to missed words

14. What is the primary function of an Embedding Layer?

Regularization
Tokenization
Handling out-of-vocabulary words
Representing words in dense vector format
Reducing sequence length

15. After training, what can be inferred if the validation loss keeps decreasing
but training loss remains high?

Model is perfectly trained

Training data is corrupted
 Model is overfitting
Model architecture is flawed
Model is underfitting

Transformers and Attention Mechanisms - Pre Quiz

1. The Transformer architecture introduced the concept of self-
attention to handle which primary challenge in sequence
modeling?

Reducing model size

Improving model robustness
Capturing dependencies regardless of their distance in the
input
Speeding up training
Handling larger input sizes

2. Why is attention particularly crucial in sequence-to-sequence

tasks like translation?

It speeds up the training process

It makes the model more interpretable
It allows the model to focus on relevant parts of the input
when producing an output
It ensures the output is of a fixed size
It reduces the model's size

3. What is the primary advantage of pretraining a Transformer on

a large corpus before fine-tuning on a specific task?

It reduces the risk of overfitting

 It allows the model to leverage general language
understanding
It makes the model smaller
It makes the model more robust to adversarial attacks
It speeds up the fine-tuning process

4. What is the primary component of the Transformer architecture

that helps it handle sequences?

Attention Mechanism
LSTM
CNN
None of the options given
RNN

5. Which Transformer model is specifically designed for language

translation?

GPT
DALL·E
BERT
T5
Image GPT

6. What does the Multi-head attention mechanism in

Transformers help with?

Improving regularization
Speeding up training
Reducing model size
Capturing different types of information from the input
 None of the options given

7. Which of the following is NOT a sequence-to-sequence task?

Translation
Question Answering
None of the options given
Summarization
Image Classification

8. What is the first step in training a Transformer model for a

specific task?

Fine-tuning
Pre-training
None of the options given
Initialization
Backpropagation

9. Which application showcases the use of Transformers in image

tasks?

Sequence alignment
Named entity recognition
Image generation using DALL·E
Text summarization
Speech recognition

10. In the context of Transformers for language translation, what

does the encoder primarily focus on?
 Reducing the sequence length
Decoding the target language
Handling attention mechanisms
Processing and representing the source language
Generating the final translation

Transformers and Attention Mechanisms –PostQuiz

1. In the context of Transformers, what does "seq to seq" stand
for?

None of the options given

Sequential to Sequential
Sequence training
Sequence to Sequence
Sequential training

2. For which task might you use a Transformer to generate a

concise summary of a long article?

Image Classification
Question Answering
None of the options given
Summarization
Translation

3. Which model can be used for both image and text tasks?

T5
GPT
None of the options given
 BERT
DALL·E

4. Which of the following models is designed for image

generation?

DALL·E
T5
GPT
None of the options given
BERT

5. Which mechanism allows Transformers to weigh the

importance of different words in a sequence?

Self Attention Mechanism

LSTM cells
None of the options given
CNN layers
RNN cells

6. In sequence-to-sequence tasks, why is attention important?

It helps the model focus on relevant parts of the input

It simplifies the model
It reduces overfitting
All of the options given
It speeds up computation

7. What is the primary task BERT is designed for?

 Text generation
Bidirectional understanding of text
Image generation
Language translation
None of the options given

8. How does Multi-head attention differ from standard attention?

It is faster
It is only used in GPT
It allows the model to focus on multiple parts of the input
simultaneously
It uses fewer parameters
None of the options given

9. Which Transformer model is known for generating coherent

paragraphs of text?

DALL·E
BERT
Image GPT
GPT
T5

10. What is the main difference between pre-training and fine-

tuning in Transformers?

None of the options given

Both are done simultaneously
Pre-training is on a large corpus and fine-tuning is task-
specific
 Fine-tuning is done without labeled data
Pre-training uses smaller models

Case Study - Transformers in Machine Translation – Quiz

1. How did Transformers improve GlobeTech's scalability issue
for new languages?

Implemented rule-based systems

Leveraged pre-trained models like BERT and GPT
Used Gradient Boosting
Introduced LSTM
Introduced RNNs

2. What unique mechanism in Transformers aids in understanding

context?

Backpropagation
CNN layers
Self-attention
Dropout
LSTM cells

3. How did the processing capabilities of Transformers affect

GlobeTech's translation time?

Had no effect
Reduced it drastically
Made it much longer
Made it slightly faster
Increased server costs
4. What technology does GlobeTech plan to integrate with
Transformers for customer support in the future?

Image recognition
Voice recognition
Gesture recognition
Augmented reality
Text summarization

5. Why can we say that Transformers brought a paradigm shift in

machine translation?

They made MT completely manual

They introduced new hardware requirements
They integrated voice translations into all platforms
They made translations context-aware and faster
They changed the way websites were designed

6. How did GlobeTech offer real-time customer support in multiple

languages?

Integrating Transformer-based MT into their chatbots

Using rule-based translations
By hiring multilingual agents
Using CNNs
Using Recurrent Networks

7. The attention mechanism in Transformers allows the model to

focus on what?
 Different parts of the input sentence
The beginning of the input sentence
The middle part of the input sentence
The graphics embedded in the text
Different parts of the output sentence

8. Which paper introduced the Transformer architecture?

"Attention Is All You Need"

"Learning Deep Architectures"
"Neural Machine Translation"
"Improving Language Understanding by Generative Models"
"Mastering the Game of Go"

9. What unique aspect is GlobeTech exploring to further enhance

translations using Transformers?

Improving voice recognition quality

Using sentiment analysis on translations
Enhancing graphics quality
Reducing translation time further
Offering translations considering regional dialects and
nuances

10. Combining voice recognition and Transformers will help

GlobeTech offer what?

Music recommendations based on voice searches

Voice-activated games
Real-time voice translations for customer support
Voice reminders for products
 Voice-activated animations

11. Traditional MT models required extensive what for each new

language?

Debugging
Re-analysis
Re-programming
Refactoring
Re-training and fine-tuning

12. How did Transformers improve GlobeTech's user interface

experience for users of different languages?

By enhancing graphics
By providing real-time translations of UI elements
By adding more interactive elements
By offering more payment options
By changing the website layout

13. Why did GlobeTech's product descriptions sound off with

earlier MT models?

Struggled with contextual meaning, especially with long

sentences
They lacked interactive elements
They were too short
They had many hyperlinks
They lacked graphics
14. What was a major challenge faced by GlobeTech in their
previous MT methods?

Contextual Translation
Graphics
Interactivity
Real-time Voice Translations
Speed

15. After adopting Transformer-based MT, by how much did

GlobeTech reduce translation-related complaints?

0.1
0.5
0.2
0.4
0.3

Generative AI Applications in Key Industries:Quiz

1. Which AI model, developed by Google, is designed to engage
in open-ended conversations, often generating creative
responses to user prompts?

Google Bard
LaMDA
 Bing AI
ChatGPT
DALL-E

2. What differentiates Google Bard's data access from ChatGPT?

ChatGPT employs discriminative AI

Bard extracts real-time information
Bard has more visual capabilities
Bard is built on GPT-4
ChatGPT offers improved visuals

3. Which AI platform, integrated into Microsoft's Bing, delivers

instant query answers?

LaMDA
Bing AI
ChatGPT
DALL-E
Google Bard

4. In the realm of manufacturing, how does generative AI impact

the design process?

By monitoring crop health

By crafting ad content
By enhancing MRI visuals
By creating product designs
By facilitating binary decisions
5. During the harvesting phase, how does AI offer a boon to the
agricultural sector?

By distinguishing inferior plants

By enhancing financial processes
By amplifying equipment resilience
By forming individual educational pathways
By translating marketing content

6. DALL-E's image generation can be optimal for which of the

following applications?

Designing book covers

Simulating cyber risk scenarios
Enhancing banking interactions
Translating ad content
Binary choice models

7. Which industry utilizes AI for personalized care programs

enhancing patient recovery?

Manufacturing
Healthcare
Cybersecurity
Advertising
Education

8. Which conversational AI is not constructed on the Transformer

neural network foundation?

ChatGPT
 Google Bard
DALL-E
Bing AI
LaMDA

9. For which sector does generative AI replicate potential threat

environments to bolster proactive defense?

Education
Finance
Advertising
Agriculture
Cybersecurity

10. Which AI methodology specializes in data set differentiation?

Generative AI
Visual AI
Discriminative AI
Binary AI
Transformer AI

Quiz: Generative AI Tools

1. Bard by Google has limitations in which of the following
aspects?

Constantly updated with web information

Programming and software development capabilities
Transformer model
 Ethically developed
Limited to English language

2. Which of the following is NOT an attribute of GPT-4 by

OpenAI?

Large multimodal model

Enhanced creativity and accuracy
Audio outputs
Persistent bias issues
Image and text input

3. How does Claude by Anthropic enhance its safety features?

Emphasizing creativity
Ethically developed
By accessing the web
Slack integration
Using "red team" prompts for safety

4. For which feature might users of the basic plans of Synthesia

encounter quality concerns?

Audio quality
Video resolution
Efficient basic content generation
Scripted prompts
Language integrations
5. What is a potential concern when using Code Whisperer by
AWS with open-source projects?

Potential open-source legal issues

It aligns with best practices
AWS optimization
Challenges with complex tasks
It boosts productivity with instant suggestions

6. Approximately what percentage of false positive rate does

AlphaCode by DeepMind have?

0.05
0.02
0.03
0.04
0.01

7. Which database does GitHub Copilot ground its data on?

Synthesia's scripted prompts

GPT-4 database
Anthropic's Claude database
DeepMind's AlphaCode repository
OpenAI Codex and GitHub

8. Cohere Generate primarily targets which type of content?

Conversational tone with Slack integration

Marketing and sales content
Video creation from scripted prompts
 Quick code generation via language prompts
Language inputs for image outputs

9. Compared to its predecessor, DALL-E, what is an improved

feature of DALL-E 2?

Same image resolution

Requires less purchase credits
Ethical development
Less safety protocols
Higher resolution

10. What is a unique feature of ChatGPT that distinguishes it from

Bard by Google?

Retains conversation history

No conversational history feature
Designed for human dialogue
Ethically developed
Built on LaMDA transformer model