AI AZURE 900 - BASICS

What is Microsoft Certified: Azure AI Fundamentals

and Exam AI - 900?
The Exam AI – 900: Microsoft Azure AI Fundamentals tests the foundational
knowledge of artificial intelligence, machine learning and services in Microsoft Azure
that are catered to AI and ML to solve complex problems. With Microsoft Azure
catering to over 19% of the market share, it is one of the most sought-after cloud
ecosystems and has huge potentials for technology enthusiasts and seasoned
professionals. Henceforth, Microsoft has different certification programs to showcase
and register your expertise in comparison to what the market demands the workforce
and with Exam AI – 900, Microsoft honors the candidates with Microsoft Certification
on Azure AI Fundamentals as one passes the exam.

Passing Score
The Passing Score for the Test is 700 and takes an estimated time of 80 Minutes.
The main exam is 60 minutes long with 30 minutes prior check-in from the appointed
allotted time.

Proctor
The Exam can be taken in Microsoft Test Centers or at the comfort at your home/
office. While taking the exam at home, you’ll be monitored via Webcam and Audio.
Your room will be administered from a call from Proctor during the check-in before
the exam. A more detailed article regarding Do’s and Don'ts during the exam will be
shared before exam.

How to Prepare for the AI – 900 Exam?

The best way to learn is to get hands-on experience. Microsoft Learn is a great
platform to get the resources which will enable one to get engaged with the Azure
Services dedicated to AI. If you have a decent amount of time, it would be wise to
explore those. These notes however in other hand, will give you a preview of the
learnings needed and help you get used to with the types of questions that will be
there during the exam.

In this article, we’ll go in brief about the different Azure Services dedicated for AI and
then explore the various modeled questions.
1. Basics of AI on Azure
This exam is dedicated to judge your understanding of AI and how Azure comes into
play. If you are an AI Engineer, you’ll be using these very tools to produce AI
enabled applications. The first step is to learn what Artificial Intelligence and Machine
Learning is and how Microsoft AI caters to this area.

Artificial Intelligence (AI)

Artificial Intelligence (AI) is the branch of computer science with multiple inter-
relations to various domains which refers to the creation of intelligence forms that
imitate human capabilities and behavior. Artificial intelligence was first ever coined in
1955 and was envisioned for general artificial intelligence during the initial inception
but later, progressed into domain-specific and task-based artificial intelligence.

To Learn more about Artificial Intelligence, Check out the previous article, Artificial
Intelligence Overview.
Machine Learning (ML)
Machine Learning refers to the process by which machines can be taught to learn
from data. It is the approach of teaching computer models to learn from data to make
predictions and draw out conclusions. Usually, huge data is needed to create these
models and to train them to design an effective and accurate system. Machine
Learning is a subject of AI and it is an approach to solve numerous problems. From
Computer Vision to Natural Language Processing to Analysis in Stock Market and
Healthcare, Machine Learning is everywhere. It is tremendously powerful and has a
subset under it to solve even exponentially difficult problems.

Microsoft AI
Microsoft AI is a powerful framework that enables organizations, researchers, and
non-profits to use AI technologies with its powerful framework which offers services
and features across domains of Machine Learning, Robotics, Data Science, IoT, and
many more. Learn more about Microsoft AI from the previous article, Microsoft Azure
AI Fundamentals. Some key features in Azure for AI are as follows.
 Cognitive Services
The tools are based on the human sense, for instance, computer vision tries to
mimic and copy the functionality of the human eye. Similarly, different listening
capabilities of humans are made a reality through Natural Language
Processing. Smart devices today able to listen and translate sound such as –
the Intelligent bot services.
 Personalized Models
The developers are enabled to make their own models without using the rest
APIs. Azure Machine Learning Services provides automated machine learning
model services.
 Drag and Drop
The Drap and Drop functionality is accessed with the Designer in Azure. You
can easily build an intelligent system with just drag and drop features. Inside
this, we can use a notebook or import your code in your favorite IDE to execute
our commands.

Anomaly Detection
Anomaly detection helps in finding out anomaly ie. Usually detection of errors and
unusual activities. Anomaly basically means something beyond normal. It is
commonly used in the field of Data Mining. Machine Learning can be implemented to
detect cases of an anomaly in multitudes of scenarios through supervised and
unsupervised methods. It is vigorously used to detect errors and failures, thus help to
prevent crisis scenarios.

2. Azure Machine Learning

Azure Machine Learning Studio
The Azure Machine Learning Studio is mostly dedicated to developers and data
scientists which provide a graphical user interface-based platform to construct and
use the workflows to solve Machine Learning problems through Azure services.

Machine Learning Models in Azure

Azure has a Machine Learning Studio which can perform all necessary tasks for
production-grade Machine Learning from Data selection to Model Building and
analysis for higher accuracy of the system.

The Model is trained, scored, and evaluated and with the experimental results,
different algorithms are used to provide better results with re-iterated operations.
Moreover, Regression, Classification and Clustering Models can be created with
Azure Machine Learning Designer.

Azure Machine Learning Designer

Azure Machine Learning Designer enables training and deployment of machine
learning models with Drag and Drop feature. The Machine Learning Pipelines can be
created using datasets and models with just Drag and Drop without the need of
having to write any code.
What kinds of models can Azure
Machine Learning Studio help create?
 Regression Model
Regression is a statistical approach to estimate relationship between a dependent
variable and one a more independent variable. It is a supervised learning technique
and can be used for the prediction of continuous value such as house price
prediction.

 Classification Model
Classification Model is also a supervised machine learning technique which focuses
on predicting classes and categories from inputted data. From Spam Detection, Dog
Breed Detection to Sentiment Analysis, Classification Model plays a huge role.

 Clustering Model
Clustering falls under unsupervised machine learning which focuses on grouping
entities that are similar to each other based on feature. Algorithms such a K-Means
can be used to cluster images based on colors.

Azure Machine Learning Designer enables to create all the above models with ease,
just with a few Drag and Drop. This enables even people without sound background
in Machine Learning to create ML Models and experiment with the capabilities of
Artificial Intelligence. Azure also supports, Automated Machine Learning ie. Auto
ML.

 Automated Machine Learning

Automated Machine Learning (Auto ML) refers to automating the machine learning
model development process which is mostly iterative and extremely time-consuming
which enables developers, analysts, and data scientists to build highly scalable,
efficient, and productive Machine Learning Models. Azure provides the feature of
Auto ML which makes it easier to obtain production-ready Machine Learning Models
without having to spend much time. Dozens of Models can be created and compared
at the same time with the accurate ones to be decided for usage.

To learn Automated Machine Learning, it would be easier if you experience hands-

on learning to create resources for Machine Learning in Azure itself. You can learn
on creating Cognitive Services from the previous article, How to Create a Cognitive
Service in Azure which is dedicated to it.
 Azure Cognitive Services
Azure Cognitive Services enables organizations to build cognitive intelligence for
applications with client library SDKs and REST APIs. Azure Cognitive Services
allows developers to integrate cognitive features into applications with no prior
knowledge of Machine Learning, Data Science, and Artificial Intelligence skillset.
From Computer Vision to Natural Language Processing and Conversational AI,
Azure Cognitive Services enables all kinds of applications.
Microsoft Azure provides various services that caters to different areas of AI
including Machine Learning, Anomaly Detection, Computer Vision, Natural Language
Processing, Conversational AI with focusing about Responsible AI in practice.

3. Computer vision
Computer Vision is synonymous with its name. This branch of AI aids by supporting
computers to analyze data from images, cameras, videos, and other visual content.
Check out object detection implemented with a hands-on example on this previous
article Object Detection with ImageAI. Computer Vision mainly deals with the
processing of visual data such as images or videos. Multitudes of Machine Learning
Models can be implemented using various Algorithms to perform different tasks.
Some of the key tasks are performed in Computer Vision are listed below.
 Face Detection
Face detection as the name suggests in a technology using which human faces can
be detected in images, videos, and other digital forms. This has huge implications for
systems from airports to shopping malls. Its applications and usage are only limited
to human imagination.

 Face Recognition
Face Recognition is the capability of a system to recognize and distinguish a specific
person through matching the digital image or video from the database of faces. From
usage for security to unlock phones to identify people in social media platforms
like Facebook, face recognition has huge implications.

To learn more with hands-on experience with Azure Cognitive Services, Check this
previous article, Face Detection and Recognition with Azure Cognitive Services.
 Custom Vision
Custom Vision in Microsoft Azure enables the training of custom image classification
and object detection models with own images.

 Object Detection
Object Detection is performed by training models to classify individual objects within
the frame of the image or video. With object detection, we can detect various images
within an image, for what the model is trained for.

 Optical Character Recognition

Optical Character Recognition (OCR) is a technique that enables the reading and
detection of text. It can be used to read characters from images, photographs,
scanned documents, currency, and more.

 Form Recognizer
With Form Recognizer Service in Microsoft Azure, information can be extracted from
the scanned invoices and forms. This feature would help create applications to
extract data with intelligent data capture eliminating manual entry minimizing human
errors.

4. Natural Language Processing

Natural Language Processing (NLP) refers to the ability of computers to understand
the natural languages of humans through audio or text means. Microsoft’s Cortana
Assistant is an example of NLP. Microsoft Azure supports to build Natural Language
Processing solutions with its Various cognitive services. They are as listed below.
Speech
The Speech service of Microsoft Azure enables developers to create applications
with the feature to synthesize and recognize speech and to translate verbal
languages.

 Language Understanding Intelligent Service

(LUIS)
Text-based commands and spoken language can be trained using the LUIS feature
provided by Microsoft Azure.

 Text Analytics
Text Analysis help to analyze text documents and to extract required key phrases,
perform sentiment analysis in term of positive of negative and supports the detection
of entities, for instance, location, people and date.

 Translator Text
The Translator Text feature provided by Microsoft Azure helps translate text in over
sixty different languages.

5. Conversational AI
Conversational AI refers to the ability of computers to engage and participate in
natural conversation in spoken or written language. Conversational AI can be
understood as this specific type of AI which is capable of having a to
and fro conversation with a human entity. It is prevalent in social media messaging
platforms, phone calls, and web interfaces to use this technology.
  QnA Maker
QnA Maker is a cognitive service provided by Microsoft AI to build a knowledge-
based question-answer system, which makes it capable for the AI agent to have a
decent conversation with the human agent.

 Azure Bot Service

Azure Bot Service enables developers with the bot framework to create, publish and
manage bots service such that back end services such as LUIS and QnA Maker can
be integrated into the system and be connected to various channels such as emails,
Microsoft Teams, and web chats.

6. Responsible AI
Artificial Intelligence has the potential to create a huge impact in almost every field of
society. It can transform industries just like electricity. From manufacturing, security,
communication, healthcare, agriculture – it can touch the lives of people across the
globe. With great power, comes great responsibility. Thus, it is crucial, Artificial
Intelligence that we create as developers are crafted thoughtfully and responsibly.
Microsoft has pointed our six different principles to abide by as the guidelines while
designing our Artificial Intelligent Systems.

The six main principles of Responsible AI are as listed below.

 Fairness
AI system should be fair to every entity. It cannot prefer one entity over the other
based on ethnicity, gender, nationality, or other superficial factors. This fairness if not
taken into care, can create havoc in judicial decision-making scenarios such as law,
justice, and criminal case analysis among others. Unfair advantages and
disadvantages must both be taken care of while developing an AI system.

 Reliability and Safety

In critical use case scenarios such as healthcare, autonomous driving – the reliability
and safety of AI is paramount. The AI system must go through rigorous testing and
approval before it can be in use.

 Privacy and Security

AI system must be secure and it must respect the privacy of individuals. Data leaks
cannot occur in the AI system. Since AI models heavily rely on data to better
themselves, it is important that security and privacy concerns are taken care of.
Some of the points to note during designing AI with privacy and security in mind are,

 Data Origin and Lineage

 Data Use – Internal vs External
 Data Corruption Considerations
 Inclusiveness
AI should be inclusive. It must engage and empower people from wide aspects of
society. After all, AI is here to help and give better lives to humans. Multitudes of
demographics must be considered while developing an AI system such that no
single group is left out or is made to feel disempowered.

 Transparency
The AI systems that we developed should be transparent. The users of the system
should be enabled and made aware of the purpose of the system, its working
protocols, and processes, and its expected limitations. This will give a transparent
outlook to the user side so that their data are not manipulated and the decision-
makings supported by the AI system in different use cases are indeed transparent.

 Accountability
AI systems should be designed with Accountability in mind. The developers and
architects of the AI system need to work within the framework of governance and
organizational principles such that the system can be relied upon by every user. The
solutions that are provided must be ensured to be ethical and up-handle the legal
code of standards.

Model Question
Model Questions for AI – 900 Exam

1. You need to build an app that will read your newspaper aloud to support users
who have reduced vision. Which service in Azure should you use?

A. Text Analytics

B. Translator Text

C. Speech

D. Language Understanding (LUIS)

2. Your website provides a feature of a chatbot which is dedicated to serve

customers. You are required to detect the emotional situation of a customer for
instance, happy, sad, neutral based on the input typed by the customer in the
chatbot. Which type of AI workload should you use?

A. Anomaly Detection

B. Semantic Segmentation

C. Regression

D. Natural Language Processing

3. Complete the Workload Type as per the column on the left to its scenario on the
right. Each workload type may be used once, more than once, or not at all.
4. Complete the Principle as per the column on the left to the description on the right.
Each principle type may be used once, more than once, or not at all.

5. Choose the three out of six guiding principles for responsible AI that Microsoft
abides by? Each correct answer presents a complete solution.

A. Knowledgeability

B. Decisiveness

C. Inclusiveness

D. Fairness

E. Opinionatedness

F. Reliability and Safety

6. What example of computer vision would be creating a bounding box around

the vehicle identified in an image?

A. Image Classification

B. Object Detection

C. Semantic Segmentation

D. Optical Character Recognition (OCR)

7. You are creating an app that can count the total number of people in a
stadium. You need to ensure their face is visible. What should you use to analyze
the images?

A. the Verify operation in the Face service

B. the Detect operation in the Face service

C. the Describe Image operation in the Computer Vision service

D. the Analyze Image operation in the Computer Vision service

8. A training dataset and validation dataset from an existing dataset needs to be

created.
Which module in the Azure Machine Learning designer would you suggest should be
used?

A. Select Columns in Dataset

B. Add Rows

C. Split Data

D. Join Data

9. Complete the Learning Type as per the column on the left to the description on the
right. Each Learning Type may be used once, more than once, or not at all.

10. Complete the Task as per the column on the left to the description on the right.
Each Learning Type may be used once, more than once, or not at all

11. Which type of Machine Learning should be used to predict the total revenue
generation for the next year looking at past data of 10 years?
A. Classification

B. Regression

C. Clustering

D. Anomaly Detection

12. Choose the Yes/ No Radio Button for each of the statements.

Answers

1. C

Explanation: The Text to Speech feature of Cognitive Services of Microsoft Azure

helps to bring apps to life with natural sounding voices. The reading of newspaper
from digital format text would be transformation of text to speech and thus, the use of
Speech which is Answer C.

2. D

Explanation: The understanding of Emotional Situation can be processed by the

Sentiment Analysis in amalgamation of language detection, topic detection and key
phrase extraction. All of these are tasks performed by Natural Language Processing.

3. These answers can be easily be decided once you’ve read this article from top to
bottom. Check into the various segments of Computer Vision, Natural Language
Processing, Conversational AI with its sub topics discussed above.
4. These answers can be easily be decided once you’ve read this article from top to
bottom. Check into the various principle under the Responsible AI to learn more that
are discussed above.

5. C, D and F. This would be a multiple choice question with check boxes to choose.

Explanation: We know, the six principles of Responsible AI are Fairness, Reliability

and Safety, Privacy and Security, Inclusiveness, Transparency and Accountability.
Each of these principles are discussed in detail above in the article.

6. B

Explanation: Among different sub topics on Computer Vision above, we discussed

about object detection. Object Detection is performed by training models to classify
individual objects within the frame of the image or video. With object detection, we
can detect various images within an image, for what the model is trained for. For
creating a bounding box around a vehicle, the vehicle should be detected in the
image which is an example of object detection. The image below would help

visualize what the question asked.

7. B

Explanation: The Face Service of Azure can provide multitudes of usage. However, it
needs to be understood the difference between Face Detection and Face
Recognition. The Question clearly asked only about counting the people whose
faces are visible which is the problem of face detection. Face detection as the name
suggests in a technology using which human faces can be detected in
images, videos, and other digital forms. This has huge implications for systems from
airports to shopping malls. Its applications and usage are only limited to human
imagination. Face Recognition on the other hand is the capability of a system to
recognize and distinguish a specific person through matching the digital image or
video from the database of faces. From usage for security to unlock phones
to identify people in social media platforms like Facebook, face recognition has huge
implications. Thus, the Detect operation of Face Service is used.

8. C

Explanation: In Machine Learning, an existing dataset should be pre-split for training

and later for validating and testing to configure the accuracy. This is essential so that
the accuracy of the system is not tampered with. If training and validating is done in
same data without splitting earlier, the accuracy would come out higher than what
really is, which is unethical. Thus, The Split Data feature in the Azure Machine
Learning Studio should be used.

9. First and foremost, we need to understand what Regression, Classification and

Clustering is. Regression is a statistical approach to estimate relationship between a
dependent variable and one a more independent variable. It is a supervised learning
technique and can be used for the prediction of continuous value such as house
price prediction. Classification Model is also a supervised machine learning
technique which focuses on predicting classes and categories from inputted data.
From Spam Detection, Dog Breed Detection to Sentiment Analysis, Classification
Model plays a huge role. Clustering falls under unsupervised machine learning which
focuses on grouping entities that are similar to each other based on feature.
Algorithms such a K-Means can be used to cluster images based on colors.

When we can have learnt what each of these learning types mean, the approach to
solving the problem can be identified with ease. The numerical value prediction for
minutes that is caused by Snowfall is a linear relationship with dependent and
independent variables, which is Regression.

The description itself describes segmenting customers into different groups. The
grouping itself clearly explains that, Clustering is the approach to choose here. The
answer is always hidden in the question itself. With proper reading and analyzing,
the right option can be chosen with practice.

The prediction of Simple – Yes / No Question is a Classification Problem. Similar to if

a dog falls on one breed or other, prediction if a student will complete a university
course would be a Classification Problem.

10. This Drag and Drop Question will feel easier once you have the technical
understanding of what each Learning Types mean. The answer to this question is
following.

Confusion Matrix
In most supervised learnings, the visualization of the performance of the algorithms
is done using a confusion matrix which displays the error matrix based on statistical
classification. It gives value to the level of confusion in the data analysis. In Azure,
the Model Evaluation module outputs the matrix with various numbers for ROC,
Precision/Recall, Lift Curves and Mainly the True Positive, True Negative, False
Positive and False Negative Value.

Feature Engineering

Feature Engineering refers to the process of using data and its specific domain
knowledge to support Machine Learning algorithms learn in a better way. Splitting
the date into months, days and year helps create features from this raw data. Its
looking confusing how the timestamp alone can be no use at times for Machine
Learning, but once divided into different parts, each individual unit of the splitted data
can be used for creation of new features. Holiday Dates, Working Dates can be
figured out once the data is split.

Feature Selection

Feature Selection is the process of selecting features for building analytical models
from the subsets of relevant data. Temperature and Pressure in this question refers
to similar feature selection in order to train the weather model.

11. B. Regression.

Explanation: This is a question of predicting a numerical value. With the past data of
10 years, a regression model can be designed to predict what value can be
extrapolated upon for the next year. Thus, Regression is the answer of this question.
Whenever, relationship can be established between dependent variable ie. A
numerical outcome and other independent variables, Regression will help us achieve
our prediction goal.

12. The answers are as shown below with the Green Boxes.

It has been clearly discussed above in this article, the price of house prediction is a
Regression Question as the independent and dependent variables relationship can
be developed through it.

Anomaly detection instead helps in finding out anomaly ie. Usually detection of
errors and unusual activities. Anomaly basically means something beyond normal
such as in detection of errors and failure, fraudulent system detection and so on.
Thus, suspicious looking sign-ins that are deviated from trended patterns can be
found by anomaly detection.

Furthermore, the prediction of patient developing diabetes is a Yes/No question

which falls under Classification problem rather than Anomaly Detection. Algorithms
such as Naïve Bayes would provide a higher accuracy result for this kind of problem.
Conclusion

Thus, in this article we went through a thorough learning for Microsoft Certified:
Azure AI Fundamentals Certification through the AI 900 Exam. We covered on what
this exam really is, how this certification can open doors for prospective candidates,
a rough pre-info about the exam from scheduling to about passing scores and the
details about the Proctor in brief before taking the exam. From there onwards, we
discussed about the process to learn and prepare for the AI-900 Exam and the
relevant resources on Microsoft Learn. Then, we discussed in brief about the various
topics that are covered in Exam AI – 900 and supported you with an overview of
various knowledge necessary for the exam from Basics of AI on Azure, Azure
Machine Learning, various types of Machine Learning and features supported by
Azure such as Computer Vision, Natural Language Processing and Conversational
AI. Thereafter, we also discussed about Responsible AI and its principle. At last, 12
Model Questions for AI – 900 Exams were discussed. These questions are very
similar to what the exam would be like and what one can expect. Model Questions
were listed above with the Answers and Explanations were described in details for
why we opted for the best suitable answers from the multiple choices. Hope you
loved this article and wish you readers the very best for your upcoming exam.