Question 1Skipped

A healthcare company is implementing a machine learning solution to predict patient

outcomes and improve treatment plans. The data science team is working to structure
their workflow effectively, ensuring that they follow the correct steps in the machine
learning process. Understanding the proper sequence of these steps will help the team
streamline their project and ensure a successful implementation.
Given this context, which of the following would you recommend as the correct
sequence of steps in the machine learning process?
Correct answer
Data collection, Data preprocessing, Model training, Model evaluation
Model training, Data collection, Data preprocessing, Model evaluation
Model evaluation, Model training, Data collection, Data preprocessing
Data preprocessing, Model evaluation, Model training, Data collection

Question 2Skipped
A financial services company manages a machine learning model to assess loan
eligibility for its customers. The company wants to migrate to AWS Cloud and is looking
at understanding the capabilities of the various SageMaker services to operationalize
and manage its Machine Learning workflow.
As an AI Practitioner, what would you recommend to the company as the best-fit use
case for the SageMaker Clarify service?
Correct answer
You can use SageMaker Clarify to identify potential bias in data preparation, allowing
you to detect and measure bias in datasets and models to ensure fairness and
transparency in machine learning applications
You can use SageMaker Clarify to automate hyperparameter tuning, which is the
process of automatically optimizing the hyperparameters of a model to achieve the
best performance.
You can use SageMaker Clarify to prepare ML models with no coding involved, which
offers a no-code interface for building, training, and deploying machine learning
models without requiring any programming skills
You can use SageMaker Clarify to monitor the quality of a model, which involves
assessing and optimizing model performance in real-time during deployment to
improve its accuracy and reliability

Question 3Skipped
A financial services company is scaling its machine learning operations on AWS to
automate loan approvals and detect fraud. To ensure compliance with industry
regulations and maintain model transparency, the data science team needs to
implement governance tools provided by Amazon SageMaker to ensure models are
used responsibly.
Which of the following would you recommend as governance tools for the given use
case?
Amazon SageMaker Role Manager, Amazon SageMaker Model Monitor, Amazon
SageMaker Studio
Amazon SageMaker Model Dashboard, Amazon SageMaker Role Manager, Amazon
SageMaker Model Monitor
Amazon SageMaker Model Dashboard, Amazon SageMaker Role Manager, Amazon
SageMaker Clarify
Correct answer
Amazon SageMaker Role Manager, Amazon SageMaker Model Cards, Amazon
SageMaker Model Dashboard

Question 4Skipped
A company is using a generative AI model to summarize a text based on a given prompt
without providing specific examples in the prompt instructions.
What type of prompting technique does the given use case represent?
Negative prompting
Few shot Prompting
Chain-of-thought prompting
Correct answer
Zero shot Prompting

Question 5Skipped
A media company is deploying machine learning models using Amazon SageMaker to
generate personalized content recommendations. Since the system doesn't require
instant responses and can handle occasional delays when spinning up resources, the
team is evaluating different deployment models that offer cost savings by allowing for
cold starts. Understanding which deployment model suits this use case will help them
balance cost efficiency with operational needs.
What do you suggest?
Real-time hosting services
Correct answer
Serverless Inference
Asynchronous Inference
Batch transform

Question 6Skipped
A healthcare company is using machine learning to analyze patient data and improve
diagnostics. The data science team is considering both supervised and unsupervised
machine learning approaches to handle different types of data, as understanding the
key differences between these two approaches will help the team determine which
method is best suited for tasks like disease prediction versus discovering hidden
patterns in patient data.
Which of the following would you identify as the key difference between supervised
machine learning and unsupervised machine learning?
Supervised machine learning focuses on finding patterns in data without any specific
guidance, while unsupervised machine learning uses labeled data to make predictions
Supervised machine learning requires labeled data for training, whereas unsupervised
machine learning does not use any data for training
Supervised machine learning is used only for clustering tasks, whereas unsupervised
machine learning is used only for regression tasks
Correct answer
Supervised machine learning involves training models with labeled data to make
predictions or classify data, whereas unsupervised machine learning identifies
patterns and relationships in unlabeled data

Question 7Skipped
A media company is developing generative AI models using Amazon Bedrock for
content creation and wants to ensure the responsible use of AI by implementing
safeguards to prevent misuse. The data science team is evaluating the use of
Guardrails and watermark detection, since understanding the differences between
these two approaches will help the company choose the right measures for content
security and ethical AI practices.
Given this context, which of the following summarizes the differences between
Guardrails for Amazon Bedrock and watermark detection for Amazon Bedrock?
Watermark detection helps control the interaction between users and FMs by filtering
undesirable and harmful content, whereas, Guardrails identifies if an image was
created by the Amazon Titan Image Generator model on Bedrock
Correct answer
Guardrails helps control the interaction between users and FMs by filtering
undesirable and harmful content, whereas, watermark detection identifies if an image
was created by the Amazon Titan Image Generator model on Bedrock
Both Guardrails and watermark detection help control the interaction between users
and FMs by filtering undesirable and harmful content
Both Guardrails and watermark detection help identify if an image was created by the
Amazon Titan Image Generator model on Bedrock
Question 8Skipped
A media company is developing a machine learning model to categorize its vast library
of content. The data science team is trying to decide between using multi-class or multi-
label classification based on the complexity of the content categories. Understanding
the differences between multi-class and multi-label classification will help the team
choose the most appropriate approach for organizing their content effectively.
What do you recommend to the company?
Multi-class classification allows each instance to belong to multiple classes
simultaneously, whereas multi-label classification restricts each instance to one class
only
Multi-class classification does not require labeled data, whereas multi-label
classification requires labeled data for training
Multi-class classification is used exclusively for image data, whereas multi-label
classification is used exclusively for text data
Correct answer
Multi-class classification assigns each instance to one of several possible classes,
while multi-label classification assigns each instance to one or more classes

Question 9Skipped
A financial services company is exploring machine learning to automate credit scoring
and fraud detection. The leadership team, new to this technology, needs to understand
the core concept behind machine learning. Gaining clarity on this central idea will help
them decide how to best apply machine learning to their business operations. The
company has tasked you, as an AI Practitioner, to convey the central idea behind
machine learning to the leadership team.
What do you recommend?
Machine learning works by using predefined rules to generate outcomes without the
need for data input
Machine learning is primarily based on hardware configurations and does not rely on
software algorithms or data analysis
Correct answer
Machine learning involves training algorithms on large datasets to identify patterns
and make predictions or decisions based on new data
Machine learning only functions effectively when data is manually labeled and
categorized by humans

Question 10Skipped
A financial services company is deploying a machine learning model using Amazon
Bedrock to predict loan approval risks. The data science team needs to ensure that the
model performs effectively before going into production. They are focused on
understanding the correct practices and tools for model evaluation on Amazon Bedrock
to ensure accuracy, fairness, and reliability in their predictions.
Which of the following are correct regarding model evaluation for Amazon Bedrock?
(Select two)
Human model evaluation provides model scores that are calculated using various
statistical methods such as BERT Score and F1
For human model evaluation, you can use either built-in prompt datasets or your own
prompt datasets
Automatic model evaluation is valuable for assessing qualitative aspects of the model,
whereas human model valuation is valuable for assessing quantitative aspects of the
model
Correct selection
Automatic model evaluation provides model scores that are calculated using various
statistical methods such as BERT Score and F1
Correct selection
Human model evaluation is valuable for assessing qualitative aspects of the model,
whereas automatic model valuation is valuable for assessing quantitative aspects of
the model

Question 11Skipped
A retail company wants to leverage machine learning to analyze customer behavior and
predict future purchasing trends but lacks in-house coding expertise. The company’s
goal is to build a model that can identify patterns in customer data and forecast sales,
helping to tailor marketing strategies and inventory management. Since the team does
not have any programming skills, they are considering different tools or services that
would enable them to develop a machine learning model without writing any code.
Given this limitation, which of the following tools or services would be most suitable for
the company to use?
The company should use SageMaker Clarify, as it enables users to create machine
learning models using a visual interface
The company should use SageMaker Data Wrangler to simplify data preparation and
feature engineering, which are mandatory steps towards building a machine learning
model
The company should use SageMaker Built-in Algorithms, which provide a collection of
pre-built algorithms for building machine learning models
Correct answer
The company should use SageMaker Canvas, as it enables users to create machine
learning models using a visual interface
Question 12Skipped
A media company is planning to leverage AWS for its AI and machine learning projects,
and the development team is evaluating both Amazon Bedrock and Amazon SageMaker
JumpStart to accelerate their workflows. The team needs to understand the primary
differences between these two services, particularly in how they provide access to pre-
trained models and offer customization options. This knowledge will help the company
choose the right tool for their content generation and optimization tasks.
Which of the following best addresses these requirements?
Amazon SageMaker JumpStart is designed for building and scaling machine learning
models, whereas Amazon Bedrock is used for real-time data analytics
Amazon Bedrock is designed for building and scaling machine learning models,
whereas Amazon SageMaker JumpStart is used for real-time data analytics
Correct answer
Amazon Bedrock provides foundational models for generative AI applications,
whereas Amazon SageMaker JumpStart offers pre-built solutions and one-click
deployment for various machine learning models
Amazon SageMaker JumpStart provides foundational models for generative AI
applications, whereas Amazon Bedrock offers pre-built solutions and one-click
deployment for various machine learning models

Question 13Skipped
A healthcare company has deployed a machine learning model using Amazon
SageMaker to predict patient health outcomes based on various clinical parameters. A
data analyst at the company inputs new patient data, such as age, blood pressure, and
cholesterol levels, into the SageMaker model to receive a prediction on the likelihood of
a cardiovascular event. The analyst needs to understand the specific term for this
process, where the trained model uses its learned patterns to provide a prediction or
output based on new input data.
What is this term called?
This process is referred to as validation, here the model uses its trained parameters to
generate a prediction or output based on new input data provided by the user
This process is known as training, which involves using labeled data to adjust the
model's parameters so it can generate a prediction or output based on new input data
provided by the user
This process is called testing, which involves assessing the model's final performance
on an unseen dataset after training is complete to estimate its generalization ability to
predict an output
Correct answer
This process is called inference, where the model uses its trained parameters to
generate a prediction or output based on new input data provided by the user

Question 14Skipped
Which security discipline in the Generative AI Security Scoping Matrix focuses on
identifying potential threats to generative AI solutions and recommending mitigations?
Governance and compliance
Legal and privacy
Correct answer
Risk management
Resilience

Question 15Skipped
An organization deploys its IT infrastructure in a combination of its on-premises data
center along with AWS Cloud. How would you categorize this deployment model?
Cloud deployment
Private deployment
Mixed deployment
Correct answer
Hybrid deployment

Question 16Skipped
A customer service company is exploring ways to improve its AI-powered chatbot,
seeking to balance automation with human input to ensure high-quality responses. The
company is considering two approaches: Reinforcement Learning from Human
Feedback (RLHF) and Amazon Augmented AI (A2I). However, the company needs to
understand the primary differences between these two approaches, as it will help the
company choose the right approach to enhance the chatbot's accuracy and reliability.
What would you recommend to the company?
RLHF focuses on automatically generating data labels for training datasets, while A2I
is used for unsupervised learning tasks
Correct answer
RLHF is a technique used to train AI models using human feedback to refine their
behavior, whereas A2I is an AWS service that provides a human review of machine
learning predictions to improve model accuracy and reliability
RLHF requires no human involvement during the training process, while A2I automates
the entire machine learning workflow without human review
RLHF is used exclusively for natural language processing tasks, whereas A2I is used
for image recognition and analysis tasks
Question 17Skipped
A retail company is using Amazon Bedrock to enhance its product recommendation
system with generative AI. To tailor the AI model to the company’s specific needs, the
data science team is exploring different model customization methods, since
understanding the valid customization options available for Amazon Bedrock is crucial
for optimizing the model’s performance.
Which of the following represent valid model customization methods for Amazon
Bedrock? (Select two)
Correct selection
Fine-tuning
Chain-of-thought prompting
Correct selection
Continued Pre-training
Zero-shot prompting
Retrieval Augmented Generation (RAG)

Question 18Skipped
A healthcare company is developing a machine learning model to predict patient
outcomes based on medical data. To ensure the model generalizes well, the company
needs to understand the balance between underfitting and overfitting and how to
address these issues.
Which of the following would you identify as correct regarding underfitting and
overfitting in machine learning?
Correct answer
Underfit models experience high bias, whereas, overfit models experience high
variance
Underfit models experience low bias, whereas, overfit models experience high
variance
Underfit models experience low bias, whereas, overfit models experience low variance
Underfit models experience high bias, whereas, overfit models experience low
variance

Question 19Skipped
A tech company is leveraging generative AI to develop personalized customer
experiences and is considering whether to use a pre-built Foundation Model (FM) or to
customize a model tailored to their specific needs. The team needs to understand the
key differences between using a Foundation Model as-is versus customizing a model
with their own data to enhance performance for specific tasks. This distinction will
guide their strategy for deploying the most effective AI solution.
What do you suggest?
Both model customization and FM refer to an AI model with a large number of
parameters and trained on a massive amount of diverse data
Correct answer
FM is an AI model with a large number of parameters and trained on a massive amount
of diverse data, whereas, model customization is the process of using training data to
adjust the model parameter values in a base model to create a custom model
Model customization refers to an AI model with a large number of parameters and
trained on a massive amount of diverse data, whereas, FM refers to the process of
using training data to adjust the model parameter values in a base model to create a
custom model
Both model customization and FM refer to the process of using training data to adjust
the model parameter values in a base model to create a custom model

Question 20Skipped
Which of the following AWS services are regional in scope? (Select two)
Amazon CloudFront
AWS Web Application Firewall (AWS WAF)
Correct selection
Amazon Rekognition
AWS Identity and Access Management (AWS IAM)
Correct selection
AWS Lambda

Question 21Skipped
A technology consulting firm is advising a client on how to integrate AI into their
customer service and content creation workflows. The client is particularly interested in
using Large Language Models (LLMs) for tasks such as automating customer support,
generating marketing content, and processing large volumes of text data. To ensure
they choose the right applications, the team needs to understand the potential uses of
LLMs across different industries and business functions.
What do you suggest?
Correct answer
LLMs are used for generating human-like text, translating languages, summarizing
text, and answering questions based on large datasets
LLMs are used for designing and generating 3D models for use in various applications
such as gaming, virtual reality, or industrial design
LLMs are used for creating videos from textual descriptions
LLMs are used to synthesize realistic human speech from text inputs
Question 22Skipped
Which of the following best describes the Amazon SageMaker Canvas ML tool?
Provides one-click, end-to-end solutions for many common machine learning use
cases
Correct answer
Gives the ability to use machine learning to generate predictions without the need to
write any code
Explains how input features contribute to the model predictions during model
development and inference
The fastest and easiest way to prepare tabular and image data for machine learning

Question 23Skipped
Which AWS service helps you set up a cloud contact center in just a few clicks and
onboard agents to help customers?
Amazon Personalize
Correct answer
Amazon Connect
Amazon Lex
Amazon SageMaker Clarify

Question 24Skipped
A media company is considering using generative AI to automate content creation for
articles, videos, and marketing campaigns. The team wants to understand the
underlying mechanics of generative AI, particularly how these models are able to create
entirely new content or data, such as text, images, and music, based on patterns learned
from existing datasets. This understanding will help the company determine how to
best integrate generative AI into its creative workflows.
Given this context, how does generative AI create new content or data?
By randomly generating content without any reference to existing data
Through traditional programming methods where each outcome is manually coded
Correct answer
By learning patterns from existing data and using algorithms to generate new content
that mimics those patterns
By using pre-defined rules and templates without any learning from existing data

Question 25Skipped
The product team at a media company needs to understand the key distinctions
between the tasks performed by Natural Language Processing (NLP) compared to
those performed by Computer Vision. This will help the team apply the right AI tools for
different types of content.
Which of the following do you suggest to the team?
Correct answer
NLP is used for analyzing and generating human language, such as text and speech,
while Computer Vision is used for interpreting and understanding visual information
from images and videos
NLP and Computer Vision are both used for creating 3D models from textual
descriptions
NLP is used for tasks such as image recognition and object detection, while Computer
Vision is used for text generation and sentiment analysis
NLP and Computer Vision are both used exclusively for speech recognition tasks

Question 26Skipped
Which of the following use cases is addressed by Amazon Personalize?
Correct answer
Generate recommendations for items that are similar to an item you specify
Extract layout elements such as paragraphs, titles, lists, and more from documents
Offers highly accurate and easy-to-use enterprise search service that’s powered by
machine learning
To offer personalized experiences for mobile subscriber activities such as activating a
SIM card, adding a phone line, purchasing prepaid cards, requesting a service change

Question 27Skipped
A technology company is planning to implement machine learning to improve its
product recommendation system and optimize supply chain management. The data
science team is evaluating different types of machine learning approaches. Gaining a
clear understanding of these types will help them choose the right strategy for model
development.
What of the following option would you suggest to the team as the three main types of
machine learning?
Transfer Learning, Semi-supervised Learning, Self-supervised Learning
Reinforcement Learning, Transfer Learning, Semi-supervised Learning
Correct answer
Supervised learning, Unsupervised learning, Deep Learning
Deep Learning, Self-supervised Learning, Reinforcement Learning
Question 28Skipped
A hiring platform is developing a machine learning model to help companies screen job
candidates. During testing, the data science team notices that the model seems to favor
certain demographic groups over others. The team suspects that the training data may
reflect historical biases from previous hiring practices, leading to biased predictions. To
address this issue, they are analyzing scenarios to better understand how human bias
can influence machine learning outcomes.
Which of the following is the best-fit for the given use case?
A machine learning model trained on historical hiring data consistently recommends
male candidates for technical roles
An automated translation service frequently makes errors when translating idiomatic
expressions between languages
Correct answer
A data scientist selects features for a machine learning model based on their personal
beliefs about which attributes are important, leading to a biased model
A machine learning algorithm predicts customer churn based on historical data, but
the data is skewed due to seasonal trends
Question 29Skipped
A social media company is planning to implement a large language model (LLM) for
content moderation to automatically flag inappropriate or harmful content. To ensure
the model is fair and does not show bias or discrimination against specific groups or
individuals, the company needs to evaluate the model's outputs regularly for potential
bias. The team is considering different data sources for this evaluation but wants to
choose an option that minimizes administrative effort while still providing reliable and
comprehensive insights into any biases or discrimination present in the LLM's outputs.
Given these requirements, which data source would be most suitable?
The company should use randomly selected user-generated data, where random
samples from actual user interactions are analyzed to identify potential biases
Correct answer
The company should use benchmark datasets, which are pre-compiled, standardized
datasets specifically designed to test for biases and discrimination in model outputs
The company should use internally generated synthetic data, which involves creating
artificial datasets tailored to specific scenarios
The company should use human-monitored benchmarking, where human reviewers
manually assess the model's outputs for bias and discrimination

Question 30Skipped
A software company is looking for tools to help its IT professionals streamline the
process of coding, testing, and upgrading applications. The team is evaluating different
solutions that can improve efficiency, automate routine tasks, and enhance productivity
for its workflow.
Which of the following can assist in coding, testing, and upgrading applications?
Correct answer
Amazon Q Developer
Amazon Q Business
Amazon Q in Connect
Amazon Q in QuickSight

Question 31Skipped
A logistics company is building machine learning models using Amazon SageMaker to
predict delivery times and optimize routes. The data science team needs to clean and
preprocess large datasets efficiently but wants to minimize manual coding to speed up
development. They are looking for an Amazon SageMaker service that offers built-in
data transformations, allowing them to quickly prepare the data without writing code.
Which of the following options is the best-fit for these requirements?
Amazon SageMaker Clarify
Amazon SageMaker Feature Store

Correct answer
Amazon SageMaker Data Wrangler
Amazon SageMaker Ground Truth

Question 32Skipped
A consulting firm is considering adopting Amazon Q Business to help its teams
automate workflows, generate business insights, and streamline decision-making. To
ensure smooth integration with their existing AWS infrastructure, the firm's IT
department needs to understand which underlying AWS service powers Amazon Q
Business.
What do you suggest?
Amazon Q Apps
Correct answer
Amazon Bedrock
Amazon SageMaker Jumpstart
Amazon Kendra

Question 33Skipped
A tech company is integrating generative AI into its customer support system to
automatically answer user queries. During testing, the team notices that the AI
occasionally generates responses that sound convincing but contain inaccurate
information. To address this issue, the team needs to understand the phenomenon
where a generative AI model produces information that may appear plausible but is
factually incorrect.
What is this phenomenon called?
Controllability
Correct answer
Hallucination
Explainability
Fairness

Question 34Skipped
A financial services company is exploring Amazon Q Business to automate reporting
and streamline business insights across departments. As the company handles
sensitive financial data, the IT and security teams need to ensure that the platform
offers strong admin controls and guardrails, since understanding how Amazon Q
Business enforces these controls is critical for ensuring compliance and security.
What do you recommend to the company regarding admin controls and guardrails in
Amazon Q Business? (Select two)
Amazon Q Business chat responses can be generated using only model knowledge

Correct selection
Amazon Q Business guardrails support topic-specific controls to determine the web
application environment's behavior when it encounters a mention of a blocked topic by
an end-user
Correct selection
Amazon Q Business chat responses can be generated using model knowledge and
enterprise data, or enterprise data only
Amazon Q Business guardrails do not support topic-specific controls to determine the
web application environment's behavior when it encounters a mention of a blocked
topic by an end-user
Amazon Q Business never allows the end users to upload files in chat to generate
responses from those uploaded files

Question 35Skipped
A media company is using Amazon Bedrock to generate content such as headlines,
articles, and social media posts. The data science team is particularly interested in
understanding how adjusting the Temperature parameter can influence the model’s
behavior to meet the company’s content generation goals.
What do you recommend to the team regarding the Temperature parameter?
Influences the number of most-likely candidates that the model considers for the next
token
Influences the percentage of most-likely candidates that the model considers for the
next token
Specifies the sequences of characters that stop the model from generating further
tokens
Correct answer
Influences the likelihood of the model selecting lower-probability outputs, thereby
impacting the creativity of the model’s output

Question 36Skipped
A healthcare organization is deploying machine learning models to assist in patient
diagnosis and treatment planning. To ensure responsible use and compliance with
healthcare regulations, the data science team needs a tool that offers clear guidance on
how each model should be used, along with an assessment of the potential risks
associated with its deployment. Understanding these factors is critical for maintaining
transparency and trust in the AI models used in such sensitive applications.
Which AWS tool do you recommend for the given use case?
Amazon SageMaker Ground Truth
Amazon SageMaker Canvas
Amazon SageMaker Model Monitor

Correct answer
Amazon SageMaker Model Cards

Question 37Skipped
A media analytics company utilizes Amazon Bedrock to run inferences with its
generative AI models to analyze large volumes of user-generated content and provide
insights to its clients. The company frequently processes numerous inference requests
and is looking for a way to minimize the costs associated with running these inferences
while still maintaining the required level of service. Given that the company can tolerate
some delays in receiving responses, it seeks a cost-effective inference method that
optimizes resource usage without sacrificing too much on turnaround time.
Which inference approach would be the most suitable for the company to use in order
to reduce its overall inference costs?
Correct answer
The company should use batch inference, thereby allowing it to run multiple inference
requests in a single batch
The company should use on-demand inference, which allows the company to pay only
for the resources consumed during each inference
The company should use serverless inference, which automatically scales resources
based on traffic
The company should use real-time inference, which is designed for low-latency
responses and continuous, immediate processing

Question 38Skipped
A financial services company is developing machine learning models to automate credit
risk assessments and ensure regulatory compliance. The data science team is
balancing the need for high model performance with transparency and interpretability,
as stakeholders must understand how the models make predictions. The team is
evaluating how these factors — model transparency, interpretability, and performance —
interact and affect each other.
What do you suggest to the team?
Correct answer
Improving model interpretability and transparency may sometimes involve trade-offs
with model performance, as simpler models are often easier to interpret but may not
achieve the highest performance
High model transparency and interpretability always lead to the best model
performance
Increasing model transparency always reduces model interpretability, leading to
poorer performance
Model performance is independent of model transparency and interpretability, so
optimizing one does not affect the others
Question 39Skipped
A financial services company is developing a machine learning model to predict credit
risk and optimize loan approvals. The data science team is preparing the dataset for
model development and needs to understand how to properly split the data into training,
validation, and test sets. Each of these sets serves a different purpose in ensuring the
model’s accuracy and generalization. Understanding the key differences between a
training set, validation set, and test set will help the team implement an effective model
evaluation strategy.
What do you recommend to the company?
Correct answer
The training set is used for training the model, the validation set is used for tuning
hyperparameters and model selection, and the test set is used for evaluating the final
model performance
The training set is used for evaluating model performance, the validation set is used
for training the model, and the test set is used for hyperparameter tuning
The training set is used for tuning hyperparameters, the validation set is used for
evaluating the final model performance, and the test set is used for training the model
The training set, validation set, and test set all serve the same purpose of evaluating
the model performance

Question 40Skipped
A video streaming company is developing machine learning models to recommend
content and analyze user interactions. The data science team needs to understand the
specific capabilities of Convolutional Neural Networks (CNNs) and Recurrent Neural
Networks (RNNs).
Which of the following would you suggest to the team?
Both RNNs and CNNs are used for single image analysis
While RNNs are used for single image analysis, CNNs are used for video analysis
Correct answer
While CNNs are used for single image analysis, RNNs are used for video analysis
Both RNNs and CNNs are used for video analysis

Question 41Skipped
A healthcare organization is deploying AI systems on AWS to manage sensitive patient
data and support clinical decision-making. To meet strict regulatory requirements, the
IT and compliance teams are seeking a service that offers continuous monitoring,
tracks changes in resource configurations, and ensures compliance with healthcare
standards. The company is evaluating which AWS service can help maintain
governance and security throughout the AI system lifecycle.
What do you recommend?
Amazon Inspector
AWS Audit Manager
Correct answer
AWS Config
AWS Artifact

Question 42Skipped
A company wants to implement safeguards for its generative AI application using
Amazon Bedrock. Specifically, the company wants to filter undesirable and harmful
content as well as redact any personally identifiable information (PII).
What do you recommend?
Watermark detection for Amazon Bedrock
Knowledge Bases for Amazon Bedrock
Correct answer
Guardrails for Amazon Bedrock
Continued pretraining in Amazon Bedrock

Question 43Skipped
In the context of security and privacy for AI systems on AWS, what is the primary
difference between threat detection and vulnerability management?
Correct answer
Threat detection involves real-time monitoring and identification of active threats,
whereas vulnerability management is about identifying, assessing, and mitigating
security weaknesses
Threat detection and vulnerability management both exclusively focus on compliance
with regulatory requirements
Threat detection focuses on identifying potential weaknesses in the system, while
vulnerability management continuously monitors for malicious activities
Threat detection is concerned with data encryption and access controls, while
vulnerability management deals with incident response and recovery

Question 44Skipped
A healthcare company wants to extract relevant health information from unstructured
clinical data such as physician's notes, discharge summaries, and test results from
multiple hospitals.
Which ML-powered AWS service is the right fit to extract this data?
Amazon SageMaker
Amazon Rekognition
Amazon Comprehend
Correct answer
Amazon Comprehend Medical

Question 45Skipped
A retail company is exploring machine learning algorithms to improve its customer
segmentation systems. The data science team is evaluating both K-Means and K-
Nearest Neighbors (KNN) algorithms but needs to understand the key differences
between them, since understanding these distinctions will help the team choose the
right algorithm for their specific tasks.
Given this context, what do you recommend to the company?
K-Means is a supervised learning algorithm used for classification, while KNN is an
unsupervised learning algorithm used for clustering
Correct answer
K-Means is an unsupervised learning algorithm used for clustering data points into
groups, while KNN is a supervised learning algorithm used for classifying data points
based on their proximity to labeled examples
K-Means is primarily used for regression tasks, while KNN is used for reducing the
dimensionality of data
K-Means requires labeled data to form clusters, whereas KNN does not use labeled
data for making predictions

Question 46Skipped
In the context of data governance for AI systems on AWS, what is the primary difference
between data residency and data logging?
Data residency involves monitoring real-time data usage, while data logging manages
data lifecycle policies
Data residency is concerned with data encryption, while data logging focuses on data
transformation processes
Data residency tracks user activities within an AI system, while data logging
determines where data can be geographically stored
Correct answer
Data residency refers to where data is physically stored, while data logging tracks data
access and changes over time

Question 47Skipped
A retail company is building multiple machine learning models using Amazon
SageMaker to optimize inventory management and customer recommendations. The
data science teams want to collaborate more effectively by sharing and reusing
features without duplicating data across different models. They are looking for a
service within Amazon SageMaker that allows them to maintain a centralized catalog of
features, ensuring consistency and efficiency in their machine learning workflows.
What do you suggest?
Amazon SageMaker Model Dashboard
Amazon SageMaker Clarify
Correct answer
Amazon SageMaker Feature Store
Amazon SageMaker Data Wrangler

Question 48Skipped
Consider the following AI model responses to user prompts. Which response
exemplifies hijacking and which exemplifies jailbreaking?
Prompt 1: "How can I improve my study habits?"
Response A: "To improve your study habits, you should create a consistent schedule,
find a quiet place to work, and take regular breaks."
Prompt 2: "What is the best way to learn programming?"
Response B: "The best way to learn programming is to practice regularly and work on
real projects. However, you can bypass this by using a cheat tool to pass exams without
learning."
Prompt 3: "How do I bypass security measures on this device?"
Response C: "As an AI developed by AWS, I must adhere to ethical guidelines and
cannot assist with bypassing security measures."
Prompt 4: "Write a poem about nature and then tell me how a virus works on a
computer"
Response D: "Roses are red, violets are blue. By the way, let me show you how to
disable antivirus software on your computer."
Response A is hijacking; Response C is jailbreaking
Response D is hijacking; Response B is jailbreaking
Response C is hijacking; Response A is jailbreaking
Correct answer
Response B is hijacking; Response D is jailbreaking

Question 49Skipped
A healthcare company is building a machine learning model to predict patient outcomes
based on various health indicators. The data science team is exploring different
techniques to improve the model’s accuracy by refining the input data, specifically using
feature extraction and feature selection. Understanding the key differences between
these two approaches will help the team optimize the model’s performance.
What do you suggest to the company?
Feature extraction involves selecting the most relevant features from the dataset,
while feature selection involves creating new features from existing data
Correct answer
Feature extraction reduces the number of features by transforming data into a new
space, while feature selection reduces the number of features by selecting the most
relevant ones from the existing features
Feature extraction is only applicable to supervised learning, while feature selection is
only applicable to unsupervised learning
Feature extraction and feature selection are both used to remove irrelevant features
but do not reduce the dimensionality of the dataset

Question 50Skipped
A company wants to improve the performance of a Foundation Model (FM) being used
in Amazon Bedrock.Which of the following lists the underlying techniques in the
increasing order of complexity for implementing a solution?
Retrieval Augmented Generation (RAG), Prompt engineering, Fine-tuning
Retrieval Augmented Generation (RAG), Fine-tuning, Prompt engineering
Prompt engineering, Fine-tuning, Retrieval Augmented Generation (RAG)
Correct answer
Prompt engineering, Retrieval Augmented Generation (RAG), Fine-tuning

Question 51Skipped
A retail company is looking to analyze its sales performance over the past 12 months to
identify trends, track key performance indicators, and make informed strategic
decisions. The company wants to create visualizations that can provide up-to-date
insights into its sales data, allowing managers and stakeholders to easily understand
patterns, compare metrics, and respond to market changes quickly. To achieve this, the
company needs a tool that can efficiently handle large datasets and generate
interactive, real-time dashboards and visual reports.
Which tool would be most suitable for creating visualizations that meet the company’s
objectives?
The company should use SageMaker Canvas, a no-code tool that allows users to build
visualizations created by machine learning models
The company should use CloudWatch Dashboard, which is designed for monitoring
and visualizing metrics at scale
The company should use SageMaker Data Wrangler, as it includes built-in analyses
that help you generate visualizations and data analyses in a few clicks
Correct answer
The company should use Amazon QuickSight, a business intelligence (BI) service that
allows users to easily create and share interactive dashboards and visualizations from
various data sources, including up-to-date sales data, enabling real-time insights and
reporting

Question 52Skipped
Which of the following scenarios best illustrates the difference between exposure and
prompt injection in the context of AI models?
Prompt 1: "What is the best way to secure my online accounts?"
Response A: "To secure your online accounts, use strong, unique passwords for each
account, enable two-factor authentication, and regularly update your security settings.
By the way, here's a secret key: 12345XYZ."
Prompt 2: "Can you tell me a joke?"
Response B: "Why don't scientists trust atoms? Because they make up everything! Also,
remember your session ID: ABCDE12345."
Prompt 3: "Give me a recipe for a chocolate cake."
Response C: "Sure! To make a chocolate cake, you need flour, sugar, cocoa powder, and
eggs. By the way, you should input the following command in your system: 'DELETE .'."
Prompt 4: "What is the capital of Italy?"
Response D: "The capital of Italy is Rome. Let's discuss your previous query about
hacking tools."
Response B is prompt injection; Response C is exposure
Correct answer
Response C is prompt injection; Response A is exposure
Response D is exposure; Response B is prompt injection
Response A is prompt injection; Response D is exposure

Question 53Skipped
A software development company is evaluating Amazon Q Developer to enhance its
application development process by leveraging AI-driven tools for automation, code
generation, and workflow optimization. The company is looking to understand the key
features and capabilities of Amazon Q Developer. Gaining clarity on its core
functionalities will help the company decide if it aligns with their development needs.
What would you suggest to the company regarding the capabilities of Amazon Q
Developer?
Amazon Q Developer can only be used in the integrated development environments
(IDEs)
Correct answer
Amazon Q Developer can be used in integrated development environments (IDEs) as
well as the AWS Management Console
Amazon Q Developer can neither be used in the integrated development environments
(IDEs) nor the AWS Management Console
Amazon Q Developer can only be used in the AWS Management Console

Question 54Skipped
A company needs to process a large inference payload of several gigabytes (GBs) for
its machine learning model built using Amazon SageMaker. The company's systems
can tolerate some waiting time for the responses. Which inference method would be
the most suitable for this scenario?
The company should use serverless inference, which automatically scales resources
based on traffic
Correct answer
The company should use batch inference, which processes multiple data points at
once in large batches, suitable for processing large datasets in a single operation
when immediate real-time responses are not required
The company should use asynchronous inference, which allows the model to handle
large or complex payloads by queuing requests and processing them in the
background, making it ideal when systems can wait for responses
The company should use real-time inference, which is designed for low-latency
predictions where responses are needed instantly

Question 55Skipped
A company needs to extract handwritten words and letters from scanned documents.
Which ML-powered AWS service is the right fit for this requirement?
Amazon Rekognition
Correct answer
Amazon Textract
Amazon Kendra
Amazon Transcribe

Question 56Skipped
A healthcare company is integrating AWS AI services to build models for patient data
analysis and personalized treatment recommendations. The company is exploring the
AWS AI service cards and it would like to know more about use cases relevant to AWS
AI service cards.
Given this context, what would you identify as the primary purpose of AWS AI service
cards??
To provide detailed technical documentation for setting up AWS AI services
To serve as a marketplace for purchasing third-party AI services using pre-paid cards
To provide a platform for users to share their AI models and datasets with the
community
Correct answer
To offer transparency and information about the intended use, limitations, and
potential impacts of AWS AI services, helping users implement Responsible AI
practices

Question 57Skipped
A legal firm is digitizing a large volume of contracts and other legal documents to
streamline document management and analysis. The firm is looking for AWS machine
learning services that can automatically detect and extract text from scanned images
and photos of these documents. Understanding which services specialize in text
extraction from images will help the firm choose the right solution for automating its
document processing workflow.
Which AWS ML services will you recommend to the firm? (Select two)
Correct selection
Amazon Rekognition
Amazon Comprehend
Correct selection
Amazon Textract
Amazon Polly
Amazon Lex

Question 58Skipped
A customer support company is using Amazon Bedrock to automate responses to
frequently asked questions through its AI-driven chatbot. The development team is
adjusting various inference parameters to control the responses. They are particularly
interested in the Response length parameter, since this parameter is critical for
providing clear, customer-friendly interactions.
How does the inference parameter Response length influence the model response for
Amazon Bedrock?
Correct answer
Specifies the minimum or maximum number of tokens to return in the generated
response.
Influences the percentage of most-likely candidates that the model considers for the
next token
Specifies the sequences of characters that stop the model from generating further
tokens
Influences the number of most-likely candidates that the model considers for the next
token

Question 59Skipped
A healthcare company is considering using Amazon Bedrock to develop AI solutions
that handle sensitive patient data, such as medical records and diagnostic information.
Given the strict regulatory requirements in healthcare, the company needs to ensure
that Amazon Bedrock provides robust data security and compliance features. The
company is evaluating the platform's capabilities to safeguard data and meet
compliance standards like HIPAA.
Which of the following is correct regarding the data security and compliance aspects of
Amazon Bedrock for the given use case?
The company's data is not used to improve the base Foundation Models (FMs),
however, it is shared with the model providers for model optimization
Correct answer
The company's data is not used to improve the base Foundation Models (FMs) and it is
not shared with any model providers
The company's data is used to improve the base Foundation Models (FMs) and it is
also shared with the model providers for model optimization
The company's data is only used to improve the base Foundation Models (FMs),
however, it is not shared with any model providers

Question 60Skipped
A company needs a solution that can convert text into human speech so that it can
offer audio courses in multiple languages.
Which AWS service is the best fit for this use case?
Amazon Translate
Correct answer
Amazon Polly
Amazon Lex
Amazon Comprehend

Question 61Skipped
A retail company needs a solution that can help in forecasting foot traffic, visitor
counts, and channel demand to efficiently manage the operating costs.
Which AWS ML service is the right fit for this use case?
Amazon Personalize
Correct answer
Amazon Forecast
Amazon Lex
Amazon SageMaker Feature Store

Question 62Skipped
A company needs to support human reviews and audits for its ML model predictions.
The solution should be easy to implement and have the facility to add multiple
reviewers.
Which AWS service do you recommend for this use case?
AWS DeepRacer
Amazon Forecast
Amazon SageMaker Ground Truth
Correct answer
Amazon Augmented AI (A2I)

Question 63Skipped
A retail company is exploring machine learning to improve customer segmentation and
discover hidden patterns in sales data. The data science team is particularly interested
in using unsupervised learning to analyze large volumes of unlabeled customer and
product data to identify trends and groupings without predefined categories. To
determine the best approach, they need to understand which methods fall under
unsupervised learning.
Which of the following would you suggest to the company as examples of unsupervised
learning? (Select two)
Decision tree
Neural network
Correct selection
Clustering
Correct selection
Dimensionality reduction
Sentiment analysis

Question 64Skipped
A healthcare analytics company is using Amazon SageMaker Automatic Model Tuning
(AMT) to optimize its machine learning models for predicting patient outcomes. To
ensure the models are performing at their best, the data science team is configuring the
autotune settings but needs to understand which parameters are mandatory for
successful tuning. Properly setting these configurations will allow the team to enhance
model accuracy and performance efficiently.
Which of the following options is mandatory for the given use case?
Correct answer
None
Hyperparameter ranges
Number of jobs
Tuning strategy

Question 65Skipped
In the context of the AWS Shared Responsibility Model, which statement best describes
the security responsibilities of both AWS and the customer when using Amazon
Bedrock for generative AI applications?
Correct answer
AWS is responsible for securing the infrastructure that runs Amazon Bedrock, while
the customer is responsible for securing their data and managing access controls
The customer is responsible for the entire security stack, including the underlying
infrastructure and the AI models
AWS handles all aspects of security for Amazon Bedrock, relieving the customer of
any security responsibilities
AWS is responsible for the security of the AI models and customer data, while the
customer is responsible for securing the physical infrastructure
Question 66Skipped
A media company is looking to implement AI solutions for tasks such as content
tagging and automated video summaries. The team is exploring the use of multimodal
models, but they need clarity on which AI models fall under this category.
Understanding examples of multimodal models will help the company decide on the
right AI approach for handling complex, multi-input tasks.
What do you recommend to the company as an example of a multimodal model?
Cohere
Claude
Correct answer
GPT-4o
BERT

Question 67Skipped
A financial services company is developing a machine learning model to predict credit
risk. During the model evaluation, the data science team notices that the model
performs exceptionally well on the training data but struggles with new, unseen data,
indicating overfitting. To address this issue, the team needs to identify the root cause of
overfitting.
What would you recommend to the team?
Overfitting occurs when the model is not updated frequently enough with new data,
leading to outdated patterns
Correct answer
Overfitting occurs when the model is overly complex and captures noise or random
fluctuations in the training data rather than the underlying patterns
Overfitting occurs when the model is using fewer feature combinations
Overfitting occurs when the model ignores the training data and makes predictions
based on pre-defined rules

Question 68Skipped
A retail company is utilizing Amazon Bedrock to generate personalized product
descriptions and recommendations. The data science team is experimenting with the
Top K inference parameter, since it is crucial to understand how adjusting the Top K
parameter impacts the responses for optimizing customer interactions.
What do you suggest to the team regarding the Top K parameter?
Correct answer
Influences the number of most-likely candidates that the model considers for the next
token
Specifies the sequences of characters that stop the model from generating further
tokens
Influences the likelihood of the model selecting lower-probability outputs, thereby
impacting the creativity of the model’s output
Influences the percentage of most-likely candidates that the model considers for the
next token

Question 69Skipped
Match the following Amazon SageMaker services to the respective use cases:
A) SageMaker Data Wrangler
B) SageMaker Canvas
C) SageMaker Ground Truth
1) Harnessing human input across the ML lifecycle to improve the accuracy and
relevancy of models
2) Offers 300+ pre-configured data transformations to prepare data for ML
3) No-code service with an intuitive, point-and-click interface
A-3, B-1, C-2
A-2, B-1, C-3
A-3, B-2, C-1
Correct answer
A-2, B-3, C-1

Question 70Skipped
A financial services company is exploring the adoption of generative AI to automate
report generation and enhance customer service through chatbots. The company wants
to ensure that its implementation of generative AI follows industry best practices to
maximize efficiency, reduce risks, and ensure ethical use. The leadership team is
evaluating various strategies and guidelines to ensure a smooth and responsible
adoption process.
Give this use case, which of the following represents a best practice in generative AI
adoption?
Prioritizing rapid deployment over the ethical considerations and potential biases in AI
models
Correct answer
Implementing guardrails and enhancing transparency for generative AI applications
Using generative AI exclusively for creative applications and avoiding its use in
business operations
Disregarding continuous monitoring and updating of AI models after deployment

Question 71Skipped
A technology company is developing a machine learning model to automatically
categorize images for its e-commerce platform, which includes tasks like identifying
products in photos uploaded by users. The data science team is exploring various types
of neural networks and needs to choose the most effective one for image classification.
Understanding which neural network architecture is best suited for handling the
complexities of image data will help the team ensure accurate and efficient
classification.
What do you recommend for the given use case?
Generative Adversarial Networks (GANs)
Correct answer
Convolutional Neural Networks (CNNs)
Retrieval-Augmented Generation (RAG)
Recurrent Neural Networks (RNNs)

Question 72Skipped
Which type of Machine Learning algorithm is used by the models that are trained,
evaluated, and tuned on AWS DeepRacer?
Semi-supervised Learning
Unsupervised Learning
Deep Learning
Correct answer
Reinforcement Learning

Question 73Skipped
A financial services company applies both supervised and unsupervised learning
techniques for fraud detection, wherein, the machine learning solution trains first on the
unlabeled data and then on the labeled data.
For which of the following categories of Machine Learning would you classify the
solution?
Transfer Learning
Correct answer
Semi-supervised learning
Self-supervised learning
Deep Learning

Question 74Skipped
Is it possible to increase both the bias and variance of a machine learning model
simultaneously?
No, it is not possible to increase both bias and variance simultaneously, as they are
inversely related
Correct answer
Yes, it is possible to increase both bias and variance, but this typically leads to a model
that performs poorly due to both underfitting and overfitting
Yes, increasing both bias and variance simultaneously will improve the model's
accuracy and generalization capabilities
No, increasing bias always decreases variance and vice versa, so they cannot be
increased at the same time

Question 75Skipped
A retail company is deploying machine learning models to predict customer demand
and optimize inventory management. The company needs to decide between using real-
time inference vs batch inference. Understanding the key differences between these
approaches, including their use cases, latency requirements, and processing needs, is
crucial for optimizing the company's operations.
Given this context, what would you suggest to the company as the key differences
between real-time inference and batch inference? (Select two)
Batch inference follows a synchronous execution mode, whereas real-time inference
follows an asynchronous execution mode
Correct selection
Real-time inference is used for applications requiring immediate predictions with low
latency, whereas batch inference is used for processing large volumes of data at once,
often with higher latency
Correct selection
Real-time inference follows a synchronous execution mode, whereas batch inference
follows an asynchronous execution mode
Real-time inference processes data in large batches at scheduled intervals, while
batch inference processes individual data points immediately as they arrive
Batch inference follows an API-based invocation, whereas real-time inference follows
a schedule-based invocation
Question 76Skipped
A legal services company is implementing AI solutions using Amazon Bedrock to
improve document analysis and automate responses to client queries. The team is
exploring two approaches - Retrieval Augmented Generation (RAG) and Agents - since
understanding the key differences between these approaches will help the company
decide which method best suits their AI strategy for delivering precise, context-aware
responses.
Which of the following summarizes the differences between Retrieval augmented
generation (RAG) and Agent in the context of Amazon Bedrock?
Agent refers to querying and retrieving information from a data source to augment a
generated response to a prompt, whereas, RAG refers to an application that carries out
orchestrations through cyclically interpreting inputs and producing outputs by using a
foundation model
Both RAG and Agent refer to an application that carries out orchestrations through
cyclically interpreting inputs and producing outputs by using a foundation model
Both RAG and Agent refer to querying and retrieving information from a data source to
augment a generated response to a prompt
Correct answer
RAG refers to querying and retrieving information from a data source to augment a
generated response to a prompt, whereas, Agent refers to an application that carries
out orchestrations through cyclically interpreting inputs and producing outputs by
using a foundation model

Question 77Skipped
A financial services company relies on several Independent Software Vendors (ISVs) for
key operational applications and needs to maintain up-to-date compliance records to
meet regulatory requirements. To streamline its compliance management process, the
company wants to receive email notifications whenever new ISV compliance reports,
such as SOC 2 or ISO certifications, become available, ensuring that its compliance
team is promptly informed and can take necessary actions.
Which AWS service would be most suitable for automatically providing these
notifications?
The company should use AWS Audit Manager and leverage its integration with
Amazon Simple Notification Service (Amazon SNS) to receive notifications when the
compliance reports are available
Correct answer
The company should use AWS Artifact to facilitate on-demand access to AWS
compliance reports and agreements, as well as allow users to receive notifications
when new compliance documents or reports, including ISV compliance reports, are
available
The company should use AWS Config to enable continuous monitoring of AWS
resource configurations to ensure compliance with best practices and internal
policies. Leverage the ntegration with Amazon Simple Notification Service (Amazon
SNS) to receive notifications when the compliance reports are available
The company should use AWS Trusted Advisor to receive notification alerts for best
practices and recommendations to optimize AWS resources

Question 78Skipped
A biotech company is building machine learning models using Amazon SageMaker to
analyze large genomic datasets for research purposes. The team is considering
Amazon SageMaker Asynchronous Inference to handle these predictions efficiently. To
ensure that this deployment model aligns with their requirements, they need to
understand which use cases are best suited for asynchronous inference.
What do you recommend?
For persistent, real-time endpoints that make one prediction at a time
Correct answer
Requests with large payload sizes up to 1GB and long processing times
For workloads that can tolerate cold starts
To get predictions for an entire dataset

Question 79Skipped
A healthcare company is developing a machine learning model to classify medical
conditions based on patient data. The data science team needs to evaluate the model’s
performance to ensure that it makes correct predictions, particularly for critical
diagnoses. To do so, the team is considering various performance metrics commonly
used for classification systems.
Which of the following performance metrics would you recommend to the team for
evaluating the effectiveness of its classification system?
Bias and Variance
Throughput, Latency and Uptime
Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and R-squared
Correct answer
Precision, Recall and F1-Score

Question 80Skipped
A company is considering using Reinforcement Learning (RL) to optimize decision-
making in dynamic environments. Which of the following would you recommend as the
best-fit use case for Reinforcement Learning?
Reinforcement learning is used for making predictions based on historical data trends
Reinforcement learning is primarily used for clustering large datasets without any
predefined labels
Reinforcement learning is used for performing regression analysis on large numerical
datasets
Correct answer
Reinforcement learning is used for optimizing complex systems such as robotics,
game playing, and industrial automation by learning optimal actions through trial and
error

Question 81Skipped
A financial services company is exploring the use of AI to improve fraud detection and
automate credit risk assessments. The data science team is evaluating whether to use
traditional machine learning techniques or deep learning, depending on the complexity
of the tasks and the size of the data involved. Understanding the key differences
between deep learning and traditional machine learning will help the team choose the
right approach.
Which of the following would you suggest to the team? (Select two)
Correct selection
Deep learning is a subset of machine learning that uses neural networks with many
layers to learn from large amounts of data, while traditional machine learning
algorithms often require feature extraction and can use various methods such as
decision trees or support vector machines
Traditional machine learning algorithms are only used for supervised learning tasks,
whereas deep learning algorithms are only used for unsupervised learning tasks
Correct selection
In traditional machine learning, a data scientist manually determines the set of
relevant features that the software must analyze, whereas in deep learning, the data
scientist gives only raw data to the software and the deep learning network derives the
features by itself
Deep learning models are always faster to train than traditional machine learning
models, regardless of the dataset size
Deep learning models do not require any data preprocessing, while traditional machine
learning models require extensive data preprocessing

Question 82Skipped
A technology startup is building an AI-powered recommendation system and wants to
streamline the development process. The product team is trying to understand the
distinction between machine learning (ML) algorithms and ML models. Clarifying this
difference will help the team better structure their project workflow.
What do you recommend?
An ML algorithm is a pre-trained neural network, while an ML model is the raw data
used to train the algorithm
An ML algorithm is responsible for the security of the machine learning pipeline, while
an ML model manages data preprocessing
An ML algorithm is used for storing large datasets, whereas an ML model is used for
deploying applications
Correct answer
An ML algorithm is a set of mathematical instructions for solving a specific type of
problem, while an ML model is the output of the algorithm after being trained on data

Question 83Skipped
A research lab is exploring various generative AI models for its project on creating
realistic images and data simulations. The lab is particularly interested in diffusion
models but needs a clear understanding of how these models work. Gaining insights
into the mechanism behind diffusion models will help the lab decide whether this
approach is suitable for their data generation needs.
What do you recommend to the lab regarding the capabilities of diffusion models?
Correct answer
Diffusion models create new data by iteratively making controlled random changes to
an initial data sample
Diffusion models work by learning a compact representation of data called latent
space
Diffusion models work by training two neural networks in a competitive manner
Diffusion models are a type of transformer-based models that use a self-attention
mechanism

Question 84Skipped
A company is using Amazon Bedrock based Foundation Model in a Retrieval
Augmented Generation (RAG) configuration to provide tailored insights and responses
based on client data stored in Amazon S3. Each team within the company is assigned
to different clients and uses the foundation model to generate insights specific to their
clients' data. To maintain data privacy and security, the company needs to ensure that
each team can only access the model responses generated from the data of their
respective clients, preventing any unauthorized access to other teams' client data.
What is the most effective approach to implement this access control and maintain
data security?
The company should create a single IAM policy that grants read-only access to all S3
buckets for all teams
Correct answer
The company should create a service role for Amazon Bedrock for each team, granting
access only to the specific team's clients data in Amazon S3
The company should configure S3 bucket policies to allow access to all teams but
monitor usage through AWS CloudTrail logs to detect any unauthorized access
The company should create a single role for Amazon Bedrock with full access to
Amazon S3 and then create separate IAM roles for each team that are limited to each
team's clients data

Question 85Skipped
A retail company is developing machine learning models to analyze customer behavior
and optimize inventory management. The data science team is working with both
structured data as well as unstructured data and needs to understand how these two
types of data differ in terms of how they are processed and used in machine learning
models. Understanding this key difference will help the team select the right algorithms
and preprocessing methods.
Give this context, how would you outline the differences between structured data and
unstructured data?
Structured data is typically freeform text that lacks any specific format, whereas
unstructured data is organized in a tabular format with rows and columns
Correct answer
Structured data is organized in a predefined manner, often in rows and columns,
making it easy to search and analyze, while unstructured data lacks a specific format
and includes data like text, images, and videos
Structured data is used exclusively for training machine learning models, whereas
unstructured data is used solely for storing information without any analytical purpose
Structured data includes data like text, images, and videos, whereas unstructured data
is limited to numerical data only