AI-900 - 공부방법 및 시험후기

1. 공부방법

1) Microsft 공식 문서 참고

• Microsoft 인증 > 학습 가이드 를 참고하여, 시험에 대한 정보, 측정 기술 등을 체크

  (※ Azure AI 서비스를 사용하여 시스템 개발을 해본 경험이 있어, MS에서 제공하는 기초적인 자기주도적 콘텐츠는 건너뛰고 Udemy의 기출문제와 인터넷에 있는 Dump 문제를 풀었다.)

2) Udemy 강좌에서 제공하는 기출문제 풀이

• AI-900: Microsoft Azure AI Fundamentals Practice Tests, 2024 에서 제공하는 5개의 실전 모의고사를 풀고 오답노트를 정리

  (해설이 제공되어 모의고사 만으로도 많은 공부를 할 수 있었다.)

3) 온라인 무료 Dump 문제 풀이

• braindumps 에서 제공하는 기출 문제를 풀었는데, 해설을 제공하지는 않는다.

---

2. 오답노트

1] 전문 개발자가 사용자 지정 Copilot을 빌드하고 Azure에서 서비스 엔드포인트로 배포하려면 어떤 도구를 사용해야 하나요?

-> Microsoft Azure AI 스튜디오

2] 생성 AI에서 퓨샷 프롬프트와 제로샷 프롬프트의 주요 차이점은 무엇인가요?

-> (정답) 퓨샷 프롬프트에서는 원하는 출력의 예를 사용하는 반면, 제로샷 프롬프트에서는 아무 것도 사용하지 않습니다. 제로샷 프롬프트와 퓨샷 프롬프트 모두 좋은 결과를 얻을 수 있음

3] What’s meaning of CRM data?

-> CRM Data refers to the information managed within a Customer Relationship Management (CRM) system. This data helps companies understand their customers and prospects more effectively, enabling personalized interactions and informed decision-making.

4] Who are the primary users of Microsoft Copilot for Sales?

-> Sellers and Sales managers

5] In which Microsoft 365 applications do Microsoft Copilot for Sales capabilities show up?

-> Teams, Outlook, and Word

6] What are the various possible channels, a bot can be connected to?

Select all correct options.

-> Alexa / Webchat / Email / Microsft teams

7] _____ is used to extract dates, quantities, and locations from text.

Select the answer that correctly completes the sentence.

-> Named Entity Recognition (NER) : Named Entity Recognition (NER) is the ability to identify different entities in text and categorize them into predefined classes or types such as: person, location, event, product, and organization.

8] You have an Azure Machine Learning pipeline that contains a Split Data module. The Split Data module outputs to a Train Model module and a Score Model module.

What is the function of the Split Data module?

-> Creating training and validation datasets : The Split Data module is particularly useful when you need to separate data into training and testing sets.

9] “Detecting a change in hospital infection rates” is a scenario of Conversational AI.

Select the correct option.

-> False : Anomaly detection (aka outlier analysis) is a step in data mining that identifies data points, events, and/or observations that deviate from a dataset’s normal behavior.

10] “Detecting spam in emails” is a scenario of Conversational AI.

Select the correct option.

-> False : Natural language processing (NLP) is the area of AI that deals with creating software that understands written and spoken language.

NLP enables you to create software that can:

o Analyze and interpret text in documents, email messages, and other sources.

o Interpret spoken language and synthesize speech responses.

o Automatically translate spoken or written phrases between languages.

o Interpret commands and determine appropriate actions.

11]

-> Verification - Do two images of a face belong to the same person?

Similarity - Does this person look like other people?

Grouping - Do all the faces belong together?

Identification - Who is this person in this group of people?

12]

-> Regression - Predict how many minutes late a flight will arrive based on the amount of snowfall at an airport.

Clustering - Segment customers into different groups to support a marketing department.

Classification - Predict whether a student will complete a university course.

13]

-> Artificial Intelligence - Any technique that enables computers to mimic human intelligence.

Machine Learning - Techniques that enable machines to improve at tasks with experience.

Deep Learning - Techniques which use layers of algorithms in the form of artificial neural networks to return results for more complex use cases.

14]

What metrics you can use to evaluate a regression model?

Select two correct options.

-> coefficient of determination (R2) is correct : R-squared (R2), or Coefficient of determination represents the predictive power of the model as a value between - inf and 1.00. 1.00 means there is a perfect fit, and the fit can be arbitrarily poor so the scores can be negative.

root mean squared error (RMSE) is correct - RMS-loss or Root Mean Squared Error (RMSE) (also called Root Mean Square Deviation, RMSD), measures the difference between values predicted by a model and the values observed from the environment that is being modeled.

15]

Which Computer Vision feature can you use to generate automatic captions for digital photographs?

Select the correct option.

-> Describe the images : Azure AI Vision can analyze an image and generate a human-readable phrase that describes its contents.

16]

You have a dataset. You need to build an Azure Machine Learning classification model that will identify defective products.

What should you do first?

-> Load the dataset.

You can think of the steps to train and evaluate a classification machine learning model as:

1) Prepare data: Identify the features and label in a dataset. Pre-process, or clean and transform, the data as needed. 2) Train model: Split the data into two groups, a training and a validation set. Train a machine learning model using the training data set. Test the machine learning model for performance using the validation data set. 3) Evaluate performance: Compare how close the model’s predictions are to the known labels. 4) Deploy a predictive service: After you train a machine learning model, you need to convert the training pipeline into a real-time inference pipeline. Then you can deploy the model as an application on a server or device so that others can use it.

17]

Which Azure Cognitive Services service can be used to identify documents that contain sensitive information?

Select the correct option.

-> Form Recognizer : Form Recognizer can analyze text in documents and forms, and with its prebuilt models, it can detect sensitive information such as personal identifiable information (PII).

18]

“Making a travel reservation” is a scenario of Conversational AI.

Select the correct option.

-> True : Making a travel reservation is an example of Conversational AI. With Microsoft’s Conversational AI tools, developers can build, connect, deploy, and manage intelligent bots that naturally interact with their users

19]

You are authoring a Language Understanding (LUIS) application to support a music festival.

You want users to be able to ask questions about scheduled shows, such as, “Which act is playing on the main stage?”

The question “Which act is playing on the main stage?” is an example of which type of element?

-> an utterance

20]

You can communicate with a bot by using email.

You can communicate with a bot by using Microsoft Teams.

You can communicate with a bot by using a webchat interface.

-> A bot can connect to multiple channels.

21]

Which statement is an example of a Microsoft responsible AI principle?

Select the correct option.

-> AI systems must be transparent and inclusive.

22]

Your team member is planning to use the Form Recognizer pre-built receipt model.

Which kind of Azure resource should you recommend to create?

-> Form Recognizer

23]

You are testing a machine learning model. How should you split data for training and evaluation?

Select the correct option.

-> Randomly split the data into some rows for training and remaining rows for evaluation.

24]

25]

Chatbot is all available.

-> A restaurant can use a chatbot to empower customers to make reservations by using a website or an app.

A restaurant can use a chatbot to answer inquiries about business hours from a webpage.

A restaurant can use a chatbot to automate responses to customer reviews on an external website.

26]

“Translating the text in an image between languages” can be performed by using the Computer Vision service.

Select the correct option.

-> False : translation text to other language require Translator APIs from Natural language processing (NLP)

27]

-> Speech - Convert a user’s speech to text.

Language service - Identify a user’s intent.

Speech - Provide a spoken response to the user.

28]

“Training of a custom image classification model” can be performed by using the Computer Vision service.

Select the correct option.

-> False : You should use Azure Custom Vision to train images using custom classification mode.

29]

What is a label and features

-> The label is the column (output) you want to predict. Features are the columns (input), which are used by the model to predict the Label.

30]

You are using Azure Machine Learning designer to build a model that will predict automobile prices.

Which type of modules should you use to complete the model?

31]

In a machine learning model, the data that is used as inputs are called ____.

Select the answer that correctly completes the sentence.

-> variables

32]

You have a solution that analyzes social media posts to extract the mentions of city names and the city names discussed most frequently.

Which type of natural language processing (NLP) workload does the solution use?

-> Entity recognition : The Named Entity Recognition module in Machine Learning Studio (classic), to identify the names of things, such as people, companies, or locations in a column of text. Named entity recognition is an important area of research in machine learning and natural language processing (NLP), because it can be used to answer many real-world questions, such as:

-Which companies were mentioned in a news article?

-Does a tweet contain the name of a person?

-Does the tweet also provide his current location?

-Were specified products mentioned in complaints or reviews?

33]

You use Azure Machine Learning designer to publish an inference pipeline. Which parameters should you use to access the web service?

Select two correct option.

-> the authentication key / the REST endpoint

34]

Automated machine learning can automatically infer the training data from the usecase provided.

Select the correct option.

-> False : AutoML generally automate the process of selecting and pre-processing the data, feature engineering, model selection, hyperparameter tuning and model evaluation, but the data needs to be provided by the user.

35]

Which of these methods is used to generate additional features?

Select the correct option.

-> Feature engineering

36]

Select two correct option for Copilot:

-> Microsoft refers “plugins developed by other companies” as third-party copilot.

Microsoft refers “Microsoft Copilot” as first-party copilot.

37]

You are building a QnA Maker bot by using a frequently asked questions (FAQ) page, and need to add professional greetings and other responses to make the bot more user friendly.

What should you do?

-> Add chit-chat : Adding chitchat to your bot makes it more conversational and engaging.

38]

During the process of Machine Learning, when should you review evaluation metrics.

Select the correct option.

-> After you test a model on the validation data.

39]

You are building a knowledge base by using QnA Maker. Which file format can you use to populate the knowledge base?

Select the correct option.

-> PDF

40]

You have an app that identifies birds in images. The app performs the following tasks:

-Identifies the location of the birds in the image

-Identifies the species of the birds in the image

Which type of computer vision does each task use?

-> Object detection, Image classification

41]

Arrange the stages in correct sequence to plan a responsible generative AI solution

Select the correct option.

-> Identify > Measure > Mitigate > Operate

Identify potential harms that are relevant to your planned solution.

Measure the presence of these harms in the outputs generated by your solution.

Mitigate the harms at multiple layers in your solution to minimize their presence and impact, and ensure transparent communication about potential risks to users.

Operate the solution responsibly by defining and following a deployment and operational readiness plan.

42]

You are building a Language Understanding model for an e-commerce business. You need to ensure that the model detects when utterances are outside the intended scope of the model.

What should you do?

-> Add utterances to the None intent

43]

“Providing an explanation of the outcome of a credit loan application” is an example of the Microsoft transparency principle for responsible AI.

Select the correct option.

-> True : Achieving transparency helps the team to understand the data and algorithms used to train the model, what transformation logic was applied to the data, the final model generated, and its associated assets. This information offers insights about how the model was created, which allows it to be reproduced in a transparent way.

44]

Which components can you drag onto a canvas in Azure Machine Learning designer?

Select two correct options.

-> dataset / module

45]

You are building an AI system and need ensure that the service meets the Microsoft transparency principle for responsible AI?.

Which task should you include ?

-> Provide documentation to help developers debug code.

46]

Which type of natural language processing (NLP) entity is used to identify a credit card number?

Select the correct option.

-> Regex entity : Regular expression entities are best for structured text or a predefined sequence of alphanumeric values that are expected in a certain format.

47]

You have been assigned an task to analyze images and text from an Image.

You are using Computer vision service to analyze image and Text Analytics service to analyze text. You have decided to use only one key and endpoint to access all of your services.

What kind of resource should you create in your Azure subscription?

-> Cognitive Services

48]

What is an example of the Microsoft responsible AI principle of transparency?

Select the correct option.

-> Helping users understand the decisions made by an AI system.

49]

Returning a bounding box that indicates the location of a vehicle in an image is an example of which computer vision capability?

Select the correct option.

-> Object detection

50]

You are using Azure ML designer to deploy a real-time inference pipeline as a service for others to consume. Where will you deploy the model?

Select the correct option.

-> AKS (Azure Kubernetes Service) : To publish a real-time inference pipeline as a service, you must deploy it to an Azure Kubernetes Service (AKS) cluster.

51]

You plan to deploy an Azure Machine Learning model by using the Machine Learning designer.

Which actions should you perform in sequence?

-> 1) Ingest and prepare dataset.

2) Split data randomly into training and validation data.

3) Train the model.

4) Evaluate model against validation dataset.

52]

-> entity recognition

53]

In the given scenario, select type of AI workload it describe.

Scenario:

An AI application that uses images to detect cracks in car windshields and warn drivers when a windshield should be repaired or replaced.

Select the correct option.

-> Computer Vision

54]

You are planning to develop a reservation system for restaurants. Which of the following services would you need?

Select two correct options.

-> Azure Bot Service / Azure QnA

55]

You need to track multiple versions of a model that was trained by using Azure Machine Learning.

What should you do?

-> Register the model : Registering the model is the correct option because it allows you to track and manage multiple versions of the model within Azure Machine Learning.

56]

-> *For a clustering model, labels must be used* is not correct. : Clustering algorithms do not use labels or target variables because the goal of clustering is to discover the underlying structure or patterns in the data, rather than to make predictions about a specific target variable. Clustering is an unsupervised learning task, meaning that the data is not labeled and the algorithm must learn to group the data into clusters based on its inherent characteristics.

57]

The first stage in a responsible generative AI process is to identify the potential harms that could affect your planned solution.

Select the four steps in this stage:

-> 1) Identify potential harms

2) Prioritize identified harms

3) Test and verify the prioritized harms

4) Document and share the verified harms

58]

“Generating live captions for a news broadcast” is an example of Speech Synthesis solution.

Select the correct option.

-> False : It would be an example of speech recognition.

59]

You are developing a model to predict events by using classification. You have a confusion matrix for the model scored on test data as shown in the following exhibit.

-> True Positive: 11 / False Negative: 1033

60]

Which all languages can you use to write custom code for Azure Machine Learning designer?

Select two correct option.

-> Python / R

61]

Which of the provided options is considered the calculated probability of a correct image classification?

Select the correct option.

-> Confidence

62]

-> *Azure Bot Service can import frequently asked questions (FAQ) to question and answer sets.* is not correct : The QnA Maker service creates knowledge base, not question and answers sets.

63]

Automated Machine Learning enabled you to specify a dataset and will automatically understand which label to predict.

Select the correct option.

-> False : The user needs to inform the system which column represents the target label, and the system will use that information to build the model.

64]

You are building a machine learning model to determine a local taxi price at a specific time of a day using historic data from a taxi service database.

Which type of machine learning model should you build?

-> Supervised learning algorithms try to model relationships and dependencies between the target prediction output and the input features such that we can predict the output values for new data based on those relationships which it learned from the previous data sets

65]

Object detection can work on identifying any defects any situation.

->Object detection can identify multiple types of damaged products in an image.

Object detection can identify the location of a damaged product in an image.

Object detection can identify multiple instances of a damaged product in an image.

66]

In which scenarios can you use speech services?

Select all correct options.

-> Providing closed captions for recorded or live videos.

Creating an automated public address system for a train station.

Creating a transcript of a telephone call or meeting.

An in-car system that reads text messages aloud.

67]

When evaluating the performance of a model, the _____ displays the predicted and actual positives and negatives by using a grid of 0 and 1 values.

Select the answer that correctly completes the sentence.

-> Confusion matrix

68]

-> *The Custom Vision service can be used to analyze video files.* is not correct : Custom Vision service can be used only on graphic files.

69]

-> *When creating an object detection model in the Custom Vision service, you must choose a classification type of either Multilabel or Multiclass* is not correct - When creating an object detection model in custom vision service, we are not provided option to select classification type.

70]

One of the stage of responsible generative AI is to Mitigate the harms at multiple layers in your solution to minimize their presence and impact, and ensure transparent communication about potential risks to users.

Select the four layers to mitigate potential harms in a generative AI solution.

-> 1) Model

2) Safety System

3) Metaprompt and grounding

4) User experience

71]

You can use Azure OpenAI services using _____

Select the answer that correctly completes the sentence.

-> Azure OpenAI Studio

72]

You need to provide content for a business chatbot that will help answer simple user queries. Validate and confirm if the below solution can be used to create question and answer text by using QnA Maker?

Solution:

Manually enter the questions and answers.

-> True

73]

You are planning to train a model, that classifies images of bananas and apples based on a collection of your own digital photographs.

Which Azure service should you use?

-> Custom Vision

74]

“An application that displays images relating to an entered search term” is an example of a conversational AI workload.

Select the correct option.

-> False : an application that displays images relating to an entered search term is not typically considered a conversational AI workload. It is an example of a task that can be performed using computer vision and natural language processing (NLP) techniques, but it is not a conversational AI workload.

75]

You need to provide content for a business chatbot that will help answer simple user queries. Validate and confirm if the below solution can be used to create question and answer text by using QnA Maker?

Solution:

Use automated machine learning to train a model based on a file that contains the questions.

-> False : You can not create question and answer text using automated machine learning, rather you can use automated machine learning to train a model based on a file that which already has question and answer text generated using QnA Maker.

**중요) chatbot은 파일 단위로 학습 시켜야 함. QnA Maker는 단순하기 때문에 수동으로 직접 QA를 입력할 수 있음

76]

-> A validation set includes the set of input examples that will be used to train a model is not correct : The validation dataset is different from the test dataset that is held back from the training of the model.

77]

To evaluate a classification model, which metric can you use?

Select the correct option.

-> true positive rate

78]

Azure Machine Learning designer enables you to include custom Javascript functions.

Select the correct option.

-> False : Azure Machine Learning designer does not enable you to include custom Javascript functions. Azure Machine Learning designer is a visual tool

79]

You need to provide content for a business chatbot that will help answer simple user queries. Validate and confirm if the below solution can be used to create question and answer text by using QnA Maker?

Solution:

Connect the bot to the Cortana channel and ask questions by using Cortana.

-> False : rather you can use the Cortana channel to connect your bot to QnA Maker and ask questions by using Cortana. To generate questions and answers in QnA Maker, you can use the QnA Maker user interface to create a new knowledge base and enter the questions and answers manually.

80]

Match the Microsoft guiding principles for responsible AI to the appropriate descriptions.

To answer, drag the appropriate principle from the column on the left to its description on the right. Each principle may be used once, more than once, or not at all.

-> Accountability : Implementing processes to ensure that decisions made by AI systems can be overridden by humans

Privacy and security : Provide consumers with information and controls over the collection, use, and storage of their data

81]

You run a charity event that involves posting photos of people wearing sunglasses on Twitter.

You need to ensure that you only retweet photos that meet the following requirements:

-Include one or more faces.

-Contain at least one person wearing sunglasses.

What should you use to analyze the images?

-> the Detect operation in the Face service

82]

A triage bot that prioritizes insurance claims based on injuries is an example of the Microsoft reliability and safety principle for responsible AI.

-> False : It is for privacy and security

83]

An AI solution that is offered at different prices for different sales territories is an example of Microsoft inclusiveness principle for responsible AI.

-> False : Not relate with inclusiveness

84]

You plan to deploy an Azure Machine Learning model as a service that will be used by client applications.

Which three processes should you perform in sequence before you deploy the model? To answer, move the appropriate processes from the list of processes to the answer area and arrange them in the correct order.

Select and Place:

-> data preparation > model training > model evaluation

85]

Data values that influence the prediction of a model are called .

-> features

86]

Automated machine learning is the process automating the time-consuming, iterative tasks of machine learning model development. is correct

87]

A banking system that predicts whether a loan will be repad is an example of the type of machine learning.

-> classification

88]

-> Automated machine learning provides you with the ability to include custom Python scripts in a training pipeline. is correct

Automated machine learning provides you with the ability to visually connect datasets and modules on an interactive canvas. is correct

89]

You are evaluating whether to use a basic workspace or an enterprise workspace in Azure Machine Learning.

What are two tasks that require an enterprise workspace? Each correct answer presents a complete solution.

-> Use a graphical user interface (GUI) to run automated machine learning experiments.

-> Use a graphical user interface (GUI) to define and run machine learning experiments from Azure Machine Learning designer.

90]

You are building a tool that will process images from retail stores and identify the products of competitors.

The solution will use a custom model.

Which Azure Cognitive Services service should you use?

-> Custom Vision

---

3. 시험후기

• PearsonView 로 원격시험 공간과 자격을 인증했는데, 인도 시험 감독관이 소리가 울리는 환경에서 환경 점검을 따로 요청했는데, 알아 듣느라 애먹었다. 다음에는 꼭 이어폰을 준비해야할 것 같다.
• 다음 단계의 자격증인 AI-102에 바로 도전하려고 한다.