A Complete Guide To Microsoft Certified - Azure AI Fundamentals (Exam AI-900)

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 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.