AI-900: Professional Analysis & Summary Report

1. Responsible AI Principles

Microsoft’s ethical framework for developing AI that users can trust.

Principle Transparency

Providing documentation and "explainability" for AI models so users understand how decisions are made.

Exam Context: Selecting "Enable Explain Best Model" in Azure ML satisfies this principle by making the "black box" of AI understandable.

Principle Inclusiveness

Ensuring AI systems empower and engage everyone, regardless of physical ability, gender, or ethnicity.

Exam Context: Identifying barriers in voice recognition for users with speech impairments is an act of supporting Inclusiveness.

2. Machine Learning Fundamentals

Core concepts used to build and evaluate predictive models.

Concept Regression vs. Classification

Regression predicts a continuous numerical value. Classification predicts a discrete label or category.

Exam Context: Predicting a taxi fare ($) is Regression; determining if a review is "Positive" or "Negative" is Classification.

Metric Confusion Matrix (TP, FP, TN, FN)

A tool to judge the accuracy of a classification model by cross-referencing predicted outcomes against actual reality.

Exam Context: A False Negative (FN) occurs when the system fails to detect a target that was actually present.

3. Computer Vision Services

Allowing software to see, identify, and process visual information.

Service Object Detection

A step beyond classification; it identifies individual items in an image and provides their exact coordinates (bounding boxes).

[Image showing the difference between image classification, object detection, and image segmentation] Exam Context: Use Object Detection to "identify the location of a damaged part" in a photo.

Service Optical Character Recognition (OCR)

The extraction of text from images, such as street signs, hand-written notes, or scanned PDFs.

Exam Context: Form Recognizer uses OCR to extract data from invoices and turn them into structured database entries.

4. Natural Language Processing (NLP)

Enabling computers to interpret, transcribe, and respond to human language.

Service Named Entity Recognition (NER)

The ability to identify and categorize key elements in text like names, dates, organizations, and locations.

Exam Context: Identifying a "phone number" or "address" within a body of email text is a function of NER.

Service Sentiment Analysis

Analyzing text to determine if the tone is positive, negative, or neutral.

Exam Context: A chatbot uses this to detect if a customer is becoming "upset" so it can escalate the chat to a human agent.

Question 2 (Page 1)

You use Azure Machine Learning designer to publish an inference pipeline. Which two parameters should you use to access the web service? [cite: 1]

A. the authentication key [cite: 3]
B. the REST endpoint [cite: 3]
C. the training endpoint [cite: 3]
D. the model name [cite: 4]

NOTE: Each correct selection is worth one point. [cite: 2]

Question 3 (Page 2)

For each of the following statements, select Yes if the statement is true. Otherwise, select No. [cite: 5, 6]

Statement	Yes	No
You train a regression model by using unlabeled data. [cite: 7]		X
The classification technique is used to predict sequential numerical values over time. [cite: 8]		X
Grouping items by their common characteristics is an example of clustering. [cite: 9]	X

Question 5 (Page 4)

You have an Azure Machine Learning model that predicts product quality. Based on the sample data below: [cite: 11]

Date	Time	Mass (kg)	Temperature (C)	Quality Test
26/02/2021	15:31:07	2.108	62.5	Pass
26/02/2021	15:31:39	2.099	62.4	Pass
26/02/2021	02:32:21	2.098	66.4	Fail

Statements	Yes	No
Mass (kg) is a feature. [cite: 15]	X
Quality Test is a label. [cite: 15]	X
Temperature (C) is a label. [cite: 16]		X

Question 8 (Page 8)

Which type of machine learning should you use to identify groups of people who have similar purchasing habits? [cite: 19]

A. Clustering [cite: 19]
B. Classification [cite: 19]
C. Regression [cite: 19]

Question 12 (Page 12)

Analyze the capabilities of Object Detection: [cite: 27]

Statement	Yes	No
Object detection can identify the location of a damaged product in an image. [cite: 29]	X
Object detection can identify multiple instances of a damaged product in an image. [cite: 30]	X
Object detection can identify multiple types of damaged products in an image. [cite: 31]	X

Question 25 (Page 25)

According to Microsoft's _________ principle of responsible AI, AI systems should NOT reflect biases from the data sets used to train them. [cite: 58]

Fairness [cite: 58]

Question 31 (Page 32)

You need to store reviews in English and present them to users in their respective language based on location. Which NLP workload should you use? [cite: 68]

A. Speech recognition [cite: 68]
B. Language modeling [cite: 68]
C. Translation [cite: 68]
D. Key phrase extraction [cite: 68]

Question 34 (Page 35)

What are three stages in a transformer model? [cite: 78]

A. Embedding calculation [cite: 78]
C. Next token prediction [cite: 78]
D. Tokenization [cite: 78]

Question 42 (Page 44)

Match tasks to machine learning models: [cite: 532]

Scenario	Model Type
Assign categories to passengers based on demographic data. [cite: 535]	Clustering
Predict the amount of consumed fuel based on flight distance. [cite: 536]	Regression
Predict whether a passenger will miss their flight based on demographic data. [cite: 531, 536]	Classification

Microsoft Azure AI Fundamentals (AI-900)

1. Responsible AI Principles

2. Machine Learning Fundamentals

3. Computer Vision Services

4. Natural Language Processing (NLP)

Azure AI Fundamentals Practice Questions