Microsoft Azure AI Fundamentals AI-900 AI-900 Questions 226–300 | Page 4/14

226

MCQmedium

A customer support team wants to automatically analyze customer emails to determine if the sentiment is positive, negative, or neutral. Which Azure service should they use?

A.Speech

B.Translator

C.Text Analytics

D.QnA Maker

AnswerC

Text Analytics includes sentiment analysis, key phrase extraction, language detection, and entity recognition, making it ideal for this task.

Why this answer

The Text Analytics service (part of Azure Cognitive Services) provides pre-built sentiment analysis, which can classify text as positive, negative, or neutral. This directly matches the requirement to automatically analyze customer emails for sentiment without needing to build custom machine learning models.

Exam trap

The trap here is that candidates may confuse Text Analytics with other NLP services like Translator or QnA Maker, mistakenly thinking any 'language' service can do sentiment analysis, but only Text Analytics has the specific pre-built sentiment model.

How to eliminate wrong answers

Option A is wrong because Azure Speech is designed for speech-to-text, text-to-speech, and speaker recognition, not for analyzing the sentiment of written text. Option B is wrong because Azure Translator focuses on machine translation between languages, not on detecting emotional tone or sentiment in the original text. Option D is wrong because QnA Maker is used to create a conversational question-and-answer layer over existing content (like FAQs), not for sentiment classification of free-form text.

Full explanation →

227

MCQmedium

What is the purpose of the Azure AI Document Intelligence's prebuilt models?

A.Training custom document extraction models for unique business forms

B.Extracting structured data from common document types (invoices, receipts, IDs) without custom training

C.Translating documents from one language to another

D.Converting documents to PDF format for archiving

AnswerB

Prebuilt models are pre-trained for common document types — you just point them at the document and receive structured extracted fields.

Why this answer

Azure AI Document Intelligence's prebuilt models are designed to extract structured data from common document types such as invoices, receipts, and IDs without requiring any custom training. They leverage pre-trained neural networks that recognize fields like invoice totals, receipt line items, and ID numbers, enabling rapid data extraction for standard forms. This aligns with the purpose of reducing manual data entry and accelerating document processing workflows.

Exam trap

The trap here is that candidates often confuse prebuilt models with custom models, assuming that all Document Intelligence models require training, when in fact prebuilt models are ready-to-use for common document types.

How to eliminate wrong answers

Option A is wrong because training custom document extraction models for unique business forms is the purpose of Document Intelligence's custom model feature, not its prebuilt models. Option C is wrong because translating documents between languages is a function of Azure AI Translator, not Document Intelligence. Option D is wrong because converting documents to PDF format for archiving is a general file conversion task, not a capability of Document Intelligence, which focuses on extracting information from documents, not changing their format.

Full explanation →

228

MCQmedium

A global news organization receives articles in multiple languages. They need to first identify the language of each article, then translate it into English. Which combination of prebuilt Azure AI services should they use?

A.Language Detection (Azure AI Language) and Translator (Azure AI Translator)

B.Key Phrase Extraction and Sentiment Analysis

C.Entity Recognition and Translator

D.Language Detection and Text Analytics for Health

AnswerA

Language Detection identifies the language of the text, and Translator converts it to the target language. Together they fulfill the requirement.

Why this answer

Option A is correct because the scenario requires two distinct steps: first, identifying the language of each article, which is performed by the Language Detection API in Azure AI Language; second, translating the article into English, which is handled by the Azure AI Translator service. These two prebuilt services are designed to work together seamlessly for multilingual content processing.

Exam trap

The trap here is that candidates may assume the Translator service alone can handle both language detection and translation, but the question explicitly requires two separate services, and the Translator's built-in detection is not a standalone prebuilt service for the first step.

How to eliminate wrong answers

Option B is wrong because Key Phrase Extraction and Sentiment Analysis are used for extracting key terms and determining the emotional tone of text, not for language identification or translation. Option C is wrong because Entity Recognition identifies named entities like people or places, but does not detect language or perform translation; Translator alone cannot detect the source language without prior language identification. Option D is wrong because Text Analytics for Health is a specialized service for extracting medical information from healthcare texts, not for general language detection or translation.

Full explanation →

229

MCQmedium

A developer uses Azure OpenAI to generate product descriptions. They provide five examples of product descriptions that follow a specific format (name, features, price, call to action). They then ask the model to write a new description for a given product, expecting the same format. Which technique is the developer using?

A.Fine-tuning

B.Zero-shot learning

C.Few-shot learning

D.Reinforcement learning

AnswerC

Few-shot learning uses a small number of examples in the prompt to demonstrate the desired output format or style, exactly as the developer does with five examples.

Why this answer

The developer is using few-shot learning, which involves providing a small number of examples (in this case, five product descriptions) to guide the model's output format and style without updating the model's weights. This technique leverages the model's in-context learning ability to follow the demonstrated pattern for a new input.

Exam trap

The trap here is that candidates often confuse few-shot learning with fine-tuning, assuming that providing examples in the prompt constitutes training the model, when in fact fine-tuning involves a separate training phase that modifies model parameters.

How to eliminate wrong answers

Option A is wrong because fine-tuning requires retraining the model on a labeled dataset to adjust its weights, which is not happening here—the developer is simply providing examples in the prompt. Option B is wrong because zero-shot learning involves no examples at all, relying solely on the model's pre-trained knowledge to generate output, whereas here five examples are explicitly given. Option D is wrong because reinforcement learning uses reward signals to iteratively improve model behavior through trial and error, not by providing static examples in a single prompt.

Full explanation →

230

MCQmedium

What is the purpose of Azure AI Translator?

A.To convert spoken audio from one language to written text in another language

B.To translate text between 100+ languages using neural machine translation

C.To detect the sentiment of multilingual text

D.To extract structured data from multilingual documents

AnswerB

Azure AI Translator provides neural machine translation for text across 100+ languages with real-time API access.

Why this answer

Azure AI Translator is a cloud-based neural machine translation service that translates text between over 100 languages and dialects. It uses deep learning models to produce high-quality, context-aware translations, making option B the correct answer.

Exam trap

The trap here is confusing Azure AI Translator with other Azure AI services that handle audio, sentiment, or document extraction, leading candidates to select options that describe related but distinct workloads.

How to eliminate wrong answers

Option A is wrong because converting spoken audio to written text in another language describes Azure AI Speech-to-Text combined with Translator, not the Translator service alone. Option C is wrong because detecting sentiment of multilingual text is a capability of Azure AI Language (specifically Sentiment Analysis), not Azure AI Translator. Option D is wrong because extracting structured data from multilingual documents is performed by Azure AI Document Intelligence (formerly Form Recognizer), not Azure AI Translator.

Full explanation →

231

MCQeasy

A robotics company is training a drone to fly autonomously through an obstacle course. The drone receives positive rewards for staying on course and avoiding obstacles, and negative rewards for collisions. The system learns by trial and error to maximize its cumulative reward. Which type of machine learning is being used?

A.Supervised learning

B.Unsupervised learning

C.Reinforcement learning

D.Semi-supervised learning

AnswerC

The drone learns by trial and error through positive and negative rewards, which is the core of reinforcement learning.

Why this answer

Reinforcement learning is the correct choice because the drone learns by interacting with its environment, receiving rewards (positive for staying on course, negative for collisions), and adjusting its behavior through trial and error to maximize cumulative reward. This is the defining characteristic of reinforcement learning, where an agent learns a policy from feedback signals rather than from labeled data or hidden patterns.

Exam trap

The trap here is that candidates often confuse reinforcement learning with supervised learning because both involve feedback, but reinforcement learning uses evaluative feedback (rewards) rather than instructive feedback (correct labels), which is the key distinction tested in AI-900.

How to eliminate wrong answers

Option A is wrong because supervised learning requires labeled input-output pairs (e.g., images with obstacle labels) to train a model, but the drone receives only reward signals, not explicit correct actions. Option B is wrong because unsupervised learning finds hidden patterns or clusters in unlabeled data without any reward or feedback, which does not match the trial-and-error reward-based learning described. Option D is wrong because semi-supervised learning combines a small amount of labeled data with a large amount of unlabeled data, but the scenario involves no labeled data at all—only reward signals from interactions.

Full explanation →

232

MCQmedium

What is 'document intelligence' (Azure AI Document Intelligence) and what types of documents can it process?

A.A service that creates documents from structured data in a database

B.A service that extracts structured data (fields, tables, key-value pairs) from forms and documents

C.A document management system for storing and organising files in Azure

D.A grammar checking tool that reviews documents for writing quality

AnswerB

Document Intelligence understands document structure to extract fields and tables from invoices, receipts, forms, and IDs.

Why this answer

Azure AI Document Intelligence (formerly Form Recognizer) is a service that uses optical character recognition (OCR) and machine learning to extract structured data—such as fields, tables, and key-value pairs—from scanned forms and documents. This enables automated processing of invoices, receipts, business cards, and other structured documents without manual data entry.

Exam trap

The trap here is that candidates confuse 'document intelligence' with general document management or editing tools, but the exam specifically tests that it is an extraction service for structured data from forms and documents.

How to eliminate wrong answers

Option A is wrong because Document Intelligence does not create documents from structured data; that describes a document generation or templating service, not an extraction service. Option C is wrong because Document Intelligence is not a document management system for storing and organizing files; that describes Azure Blob Storage or SharePoint, not an AI-based extraction service. Option D is wrong because Document Intelligence does not perform grammar checking or writing quality review; that describes a natural language processing tool like Azure AI Language's text analysis, not a document extraction service.

Full explanation →

233

MCQhard

A bank deploys an AI system that uses a complex deep learning model to approve or reject loan applications. When a loan is rejected, customers demand to know the specific reasons. The bank wants to ensure the AI system operates in a way that allows them to explain its decisions. Which Microsoft responsible AI principle is most directly relevant to this requirement?

A.Reliability and safety

B.Transparency

C.Privacy and security

D.Fairness

AnswerB

Transparency (Interpretability) ensures that AI decisions can be understood and explained, which is what the bank needs for loan rejection explanations.

Why this answer

The bank's requirement to explain why a loan was rejected directly aligns with the transparency principle, which mandates that AI systems be understandable and that their decisions can be communicated to users. In this scenario, the complex deep learning model must be interpretable, often through techniques like feature importance analysis or surrogate models, to provide specific reasons for rejection. Transparency ensures that customers can receive meaningful explanations, building trust and enabling accountability.

Exam trap

The trap here is that candidates often confuse transparency with fairness, assuming that explaining a decision automatically ensures it is fair, but transparency is solely about understandability and communication, not about the absence of bias.

How to eliminate wrong answers

Option A is wrong because reliability and safety focus on the system performing consistently and without harm (e.g., avoiding crashes or incorrect outputs), not on explaining decisions to customers. Option C is wrong because privacy and security concern protecting data from unauthorized access or breaches, not the ability to articulate the rationale behind a specific decision. Option D is wrong because fairness addresses bias and equitable treatment across groups (e.g., ensuring no discrimination based on race or gender), but does not inherently require the system to provide explanations for individual rejections.

Full explanation →

234

Drag & Dropmedium

Drag and drop the steps to create a bot with Azure Bot Service into the correct order.

Drag steps to the numbered slots on the right, or tap a step then tap a slot.

Steps

Order

Why this order

Creating a bot involves provisioning a resource, developing logic, testing, and connecting channels.

Full explanation →

235

MCQhard

A botanist uses Azure Automated Machine Learning to train a model that classifies iris flowers into three species: setosa, versicolor, and virginica. The dataset contains exactly 50 examples of each species, making it perfectly balanced. The botanist wants the primary metric to give equal importance to the classification performance of each species, regardless of their frequency. Which primary metric should the botanist select in Azure AutoML?

A.Accuracy

B.Weighted F1

C.Macro F1

D.Micro F1

AnswerC

Correct because macro F1 averages the F1 scores of all classes without weighting by class size, thereby giving equal importance to the classification performance of each species.

Why this answer

Macro F1 is the correct primary metric because it computes the F1 score independently for each class and then takes the unweighted average, giving equal importance to each species (setosa, versicolor, virginica) regardless of their balanced frequency. This aligns with the botanist's requirement to treat each species equally in performance evaluation.

Exam trap

The trap here is that candidates often confuse Macro F1 with Weighted F1, assuming that because the dataset is perfectly balanced, Weighted F1 and Macro F1 yield the same value, but the question explicitly tests the intent of 'equal importance to each species' which is the definition of Macro averaging, not the support-weighted approach.

How to eliminate wrong answers

Option A is wrong because accuracy, while simple, does not inherently give equal importance to each class; it treats all correct predictions equally and can be misleading if class imbalance existed, though here the dataset is balanced, the requirement is specifically about equal class-level importance, which accuracy does not explicitly enforce. Option B is wrong because Weighted F1 computes the F1 score for each class and averages them weighted by the number of true instances per class, which would give equal weight to each species only if the classes are perfectly balanced (which they are), but the botanist wants explicit equal importance regardless of frequency, and Weighted F1 is designed to account for support size, not to ignore it. Option D is wrong because Micro F1 aggregates contributions of all classes to compute a global metric, effectively treating each instance equally rather than each class equally, which does not satisfy the requirement of giving equal importance to each species.

Full explanation →

236

MCQmedium

What does the 'read' operation in Azure AI Vision do?

A.Reads and describes what's happening in a video

B.Extracts printed and handwritten text from images and documents

C.Reads and verifies digital signatures in documents

D.Reads metadata (EXIF data) embedded in image files

AnswerB

The Read API/OCR extracts text from images and PDFs, returning text content with spatial location information.

Why this answer

The 'read' operation in Azure AI Vision is specifically designed to extract printed and handwritten text from images and documents using Optical Character Recognition (OCR) technology. It returns the detected text along with bounding box coordinates and confidence scores, making it suitable for digitizing documents, processing forms, and extracting text from photos.

Exam trap

The trap here is that candidates confuse the 'read' operation with the 'analyze image' operation (which describes images) or assume it handles video, but the 'read' API is strictly for text extraction from static images and documents.

How to eliminate wrong answers

Option A is wrong because the 'read' operation does not analyze video content; video analysis is handled by Azure Video Indexer or the Video Analyzer service, not the 'read' API. Option C is wrong because the 'read' operation does not verify digital signatures; signature verification is a cryptographic function typically performed by Azure Key Vault or custom PKI solutions, not by computer vision OCR. Option D is wrong because the 'read' operation does not read metadata like EXIF data; EXIF data is extracted using image processing libraries or Azure Media Services, while the 'read' API focuses solely on text content within the image.

Full explanation →

237

MCQmedium

A city's traffic department wants to predict the number of cars that will cross a particular bridge each day to plan maintenance schedules. The output of the model should be a numerical value representing the estimated traffic count. Which type of machine learning task is this?

A.Classification

B.Regression

C.Clustering

D.Reinforcement learning

AnswerB

Regression predicts a continuous numeric output, which matches the requirement of estimating a traffic count.

Why this answer

Regression is the correct type of machine learning task because the goal is to predict a continuous numerical value—the number of cars crossing the bridge each day. Unlike classification, which predicts discrete categories, regression models output a real number, making it ideal for forecasting traffic counts.

Exam trap

The trap here is that candidates may confuse regression with classification because both involve prediction, but the key distinction is that regression outputs a continuous number while classification outputs a discrete label.

How to eliminate wrong answers

Option A is wrong because classification predicts discrete class labels (e.g., 'high traffic' vs 'low traffic'), not a continuous numerical value. Option C is wrong because clustering groups unlabeled data into clusters based on similarity, without predicting a specific numeric output. Option D is wrong because reinforcement learning involves an agent learning optimal actions through rewards and penalties in an environment, not predicting a single numeric value from input features.

Full explanation →

238

MCQmedium

What is 'MLOps' and how does it relate to AI workloads on Azure?

A.Operational procedures for Microsoft 365 mail system administration

B.Applying DevOps practices (automation, CI/CD, monitoring) to the machine learning lifecycle

C.A certification program for ML engineers working with Azure

D.The process of optimising ML model inference speed for production deployment

AnswerB

MLOps automates training, evaluation, deployment, and monitoring — enabling consistent, reliable ML model updates at scale.

Why this answer

MLOps (Machine Learning Operations) is the application of DevOps principles—such as automation, continuous integration/continuous deployment (CI/CD), and monitoring—to the machine learning lifecycle. On Azure, MLOps is implemented through services like Azure Machine Learning, which provides pipelines, model registries, and automated retraining to manage the end-to-end ML workflow from data preparation to deployment and monitoring.

Exam trap

The trap here is that candidates confuse MLOps with a specific technical task like model optimization (Option D) or mistake it for a certification (Option C), rather than recognizing it as the comprehensive DevOps-inspired lifecycle management practice for ML workloads.

How to eliminate wrong answers

Option A is wrong because it describes operational procedures for Microsoft 365 mail system administration, which is unrelated to machine learning operations. Option C is wrong because MLOps is a set of practices, not a certification program; Azure offers certifications like AI-900, but MLOps itself is not a certification. Option D is wrong because it refers to model optimization for inference speed (e.g., quantization or pruning), which is a specific task within the ML lifecycle, not the overarching operational framework that MLOps encompasses.

Full explanation →

239

MCQeasy

What is the 'Azure AI Language SDK' and what programming languages does it support?

A.A new programming language created by Microsoft for building NLP applications

B.Client libraries for Python, .NET, Java, and JavaScript for programmatic access to Azure AI Language

C.A software development kit for building physical language translation devices

D.An IDE plugin that adds Azure AI Language auto-complete to code editors

AnswerB

The SDK provides language-specific clients — handling auth and serialisation so developers can call NLP APIs from their preferred language.

Why this answer

The Azure AI Language SDK is a set of client libraries that allow developers to integrate Azure AI Language capabilities—such as sentiment analysis, key phrase extraction, and language understanding—directly into their applications. It supports Python, .NET, Java, and JavaScript, enabling programmatic access to the service via RESTful APIs or native SDK methods.

Exam trap

The trap here is confusing an SDK (a set of client libraries) with a new programming language or a hardware toolkit, leading candidates to incorrectly select options that describe unrelated concepts.

How to eliminate wrong answers

Option A is wrong because the Azure AI Language SDK is not a new programming language; it is a collection of client libraries for existing languages. Option C is wrong because the SDK is for software development, not for building physical hardware devices like language translation devices. Option D is wrong because the SDK is not an IDE plugin; it is a set of libraries and tools that can be used in any development environment, not limited to auto-complete features.

Full explanation →

240

MCQmedium

A hotel chain wants to automatically determine whether online guest reviews express a positive, negative, or neutral opinion about their stays. Which built-in Azure AI Language feature should they use?

A.Named Entity Recognition (NER)

B.Key phrase extraction

C.Sentiment analysis

D.Language detection

AnswerC

Sentiment analysis directly detects the positivity, negativity, or neutrality of text, which matches the requirement.

Why this answer

Sentiment analysis is the correct Azure AI Language feature because it is specifically designed to classify text as positive, negative, or neutral, which directly matches the hotel chain's requirement to determine guest opinions from online reviews. This feature uses machine learning models to evaluate the overall sentiment expressed in a document or sentence, providing a confidence score for each sentiment category.

Exam trap

The trap here is that candidates often confuse key phrase extraction with sentiment analysis, thinking that extracting phrases like 'bad service' implies sentiment, but key phrase extraction only identifies topics without evaluating their emotional polarity.

How to eliminate wrong answers

Option A is wrong because Named Entity Recognition (NER) identifies and categorizes entities like people, places, or organizations from text, not the emotional tone or opinion. Option B is wrong because Key phrase extraction identifies the main points or topics in text, such as 'clean room' or 'friendly staff', but does not classify the sentiment (positive/negative/neutral). Option D is wrong because Language detection identifies the language of the text (e.g., English, Spanish), not the sentiment or opinion expressed.

Full explanation →

241

MCQmedium

A bank uses an AI system to approve or deny personal loan applications. Several customers whose loans were denied have asked for an explanation of why their application was rejected. Which Microsoft responsible AI principle requires the bank to provide understandable reasons for the AI's decision?

A.Reliability and safety

B.Fairness

C.Transparency

D.Privacy and security

AnswerC

Transparency requires that AI systems are understandable and that decisions can be explained in meaningful terms.

Why this answer

Transparency is the Microsoft responsible AI principle that requires AI systems to be understandable and interpretable. In this scenario, the bank must provide clear, understandable reasons for loan denials, which directly aligns with transparency's goal of enabling users to understand how and why decisions are made. This principle ensures that AI outcomes are not opaque black-box decisions but can be explained in human terms.

Exam trap

The trap here is that candidates often confuse transparency with fairness, thinking that explaining a decision inherently ensures it is fair, but transparency only requires the explanation to be provided, not that the decision itself is unbiased.

How to eliminate wrong answers

Option A is wrong because reliability and safety focus on ensuring the AI system operates consistently and without causing harm, not on providing explanations for decisions. Option B is wrong because fairness addresses bias and equitable treatment across groups, but does not inherently require the system to explain its reasoning to individuals. Option D is wrong because privacy and security concern protecting data from unauthorized access and misuse, not the interpretability or explanation of AI decisions.

Full explanation →

242

MCQmedium

What is 'summarisation quality' evaluation and what metrics are used?

A.Measuring summary quality by counting how many sentences were preserved from the original

B.ROUGE scores measuring n-gram overlap between generated and reference summaries

C.Asking users to rate summary quality on a 1-10 scale in production

D.Measuring how much shorter the summary is compared to the original document

AnswerB

ROUGE-1/2/L measure different levels of overlap — the standard automatic evaluation metrics for summarisation quality.

Why this answer

Summarisation quality is evaluated using ROUGE (Recall-Oriented Understudy for Gisting Evaluation) scores, which measure the overlap of n-grams, word sequences, or word pairs between a generated summary and one or more reference summaries. This automated metric correlates well with human judgment and is standard in NLP tasks like text summarisation.

Exam trap

The trap here is that candidates confuse a simple heuristic (like length reduction or sentence preservation) with the standard automated metric ROUGE, which is specifically designed for summarisation quality evaluation in NLP.

How to eliminate wrong answers

Option A is wrong because counting preserved sentences does not capture semantic quality or conciseness; a summary could retain all sentences but still be verbose and unfocused. Option C is wrong because human rating in production is a subjective evaluation method, not a standardised automated metric like ROUGE, and it is not a defined 'summarisation quality' metric in NLP evaluation. Option D is wrong because measuring length reduction alone ignores content relevance and accuracy; a very short summary could omit critical information.

Full explanation →

243

MCQmedium

A marketing team uses Azure OpenAI Service to generate headline ideas for a campaign. They find the generated headlines are often too similar and lack creativity. Which parameter should they increase to introduce more randomness in the generated text?

A.Frequency penalty

B.Top_p (nucleus sampling)

C.Temperature

D.Presence penalty

AnswerC

Temperature directly controls the level of randomness; increasing it makes the model more likely to choose less probable tokens, leading to more creative and varied outputs.

Why this answer

Option C (Temperature) is correct because temperature controls the randomness of token selection in the model's probability distribution. Increasing temperature (e.g., from 0.7 to 1.0) flattens the probability curve, making lower-probability tokens more likely to be chosen, which introduces more diversity and creativity in the generated headlines.

Exam trap

The trap here is that candidates often confuse temperature with frequency or presence penalties, thinking that penalizing repetition (frequency penalty) will increase creativity, when in fact temperature directly controls the randomness of token selection, which is the key to generating more diverse and creative text.

How to eliminate wrong answers

Option A (Frequency penalty) is wrong because it reduces repetition by penalizing tokens that have already appeared in the text, but it does not directly increase randomness; it only discourages the model from reusing the same words or phrases. Option B (Top_p, or nucleus sampling) is wrong because it limits the cumulative probability mass of tokens considered for sampling (e.g., top_p=0.9 means only the top 90% of probability mass is sampled), which can actually reduce randomness by cutting off the long tail of low-probability tokens. Option D (Presence penalty) is wrong because it penalizes tokens that have appeared at least once in the text, encouraging the model to introduce new topics, but it does not increase randomness in token selection; it only promotes novelty in content.

Full explanation →

244

MCQeasy

What is 'key phrase extraction' in Azure AI Language?

A.Encrypting sensitive phrases in a document for secure storage

B.Identifying the most important words and phrases that best represent a text's main topics

C.Finding and extracting password-like phrases from user messages for security monitoring

D.Selecting the highest-scoring responses from a list of candidate answers

AnswerB

Key phrase extraction surfaces the key concepts in text — enabling topic summarisation, tagging, and content understanding.

Why this answer

Key phrase extraction in Azure AI Language uses natural language processing to identify the most salient words and phrases that summarize the main topics of a text. It analyzes the document's structure and semantics to return a ranked list of key phrases, enabling quick understanding of core content without reading the entire text.

Exam trap

The trap here is confusing key phrase extraction with entity recognition or extractive question answering, as all three involve extracting text but serve fundamentally different purposes—key phrases summarize topics, entities identify specific named items, and QA retrieves direct answers to questions.

How to eliminate wrong answers

Option A is wrong because key phrase extraction does not involve encryption or secure storage; it is a text analysis feature for identifying important concepts, not a security mechanism. Option C is wrong because key phrase extraction is not designed to find password-like phrases; it focuses on general topic extraction, not security monitoring or credential detection. Option D is wrong because key phrase extraction does not select from a list of candidate answers; that describes extractive question answering, a different Azure AI Language feature.

Full explanation →

245

MCQmedium

A global company receives customer support tickets in over 60 languages. They need to automatically detect the language of each ticket so it can be routed to the appropriate language-specific team. The company has no labeled training data for language identification. Which Azure AI Language feature should they use?

A.Custom Text Classification

B.Language Detection

C.Key Phrase Extraction

D.Entity Recognition

AnswerB

Language Detection is a pre-built capability that identifies the language of text without needing custom training data. It supports many languages and is ideal for this scenario.

Why this answer

Language Detection is the correct choice because it is a pre-built, zero-shot Azure AI Language feature that can automatically identify the language of text without requiring any labeled training data. The service uses a multilingual model trained on large datasets to detect over 100 languages, making it ideal for routing support tickets in over 60 languages with no prior customization.

Exam trap

The trap here is that candidates might confuse Language Detection with Custom Text Classification, assuming they need to train a model for a multilingual scenario, when in fact Azure provides a built-in, no-code language detection API that requires zero training data.

How to eliminate wrong answers

Option A is wrong because Custom Text Classification requires labeled training data to build a custom model, which the company does not have. Option C is wrong because Key Phrase Extraction identifies important terms in text but does not detect the language of the input. Option D is wrong because Entity Recognition extracts named entities like people, places, or organizations, not the language of the text.

Full explanation →

246

MCQmedium

A marketing team uses Azure OpenAI Service to generate ad copy. They notice the model sometimes uses offensive language. Which Azure OpenAI feature should they use to automatically block such content?

A.Setting the temperature parameter to 0.0

B.Using the frequency_penalty parameter

C.Enabling content filtering

D.Configuring the max_tokens parameter

AnswerC

Content filtering is a built-in safety feature designed to detect and block harmful or offensive content in both prompts and completions.

Why this answer

Option C is correct because Azure OpenAI Service includes built-in content filtering that automatically detects and blocks offensive or harmful language in both prompts and completions. This feature uses AI-based classifiers to enforce responsible AI policies without requiring manual configuration of model parameters.

Exam trap

The trap here is that candidates confuse model parameters (temperature, frequency_penalty, max_tokens) with safety features, assuming they can control content appropriateness, when in fact content filtering is a separate, dedicated mechanism in Azure OpenAI Service.

How to eliminate wrong answers

Option A is wrong because setting the temperature parameter to 0.0 controls randomness in output generation, not content safety; it makes responses more deterministic but does not filter offensive language. Option B is wrong because the frequency_penalty parameter reduces repetition by penalizing tokens that have already appeared, which has no effect on blocking offensive or harmful content. Option D is wrong because configuring the max_tokens parameter limits the length of the generated response but does not inspect or block inappropriate language.

Full explanation →

247

MCQmedium

A marketing team wants to use Azure OpenAI to generate blog post outlines. They have a single example of an outline that follows their preferred structure: introduction, three key points, conclusion. They want the model to generate new outlines that follow the same structure without retraining the model. Which technique should they use?

A.Fine-tuning the model on a large dataset of blog outlines

B.Providing the example outline in the prompt (few-shot learning)

C.Setting the temperature parameter to a high value

D.Using the Azure OpenAI embeddings API

AnswerB

Correct: Few-shot learning uses examples in the prompt to guide the model's output without retraining.

Why this answer

Option B is correct because few-shot learning involves providing a small number of examples (in this case, one example outline) directly in the prompt to guide the model's output format and structure without any retraining. This technique leverages the model's in-context learning ability to mimic the given pattern, making it ideal for generating new outlines that follow the same structure.

Exam trap

The trap here is that candidates often confuse few-shot learning with fine-tuning, assuming that any task requiring consistent output format must involve retraining the model, when in fact in-context learning via prompt engineering is sufficient for small numbers of examples.

How to eliminate wrong answers

Option A is wrong because fine-tuning requires a large, labeled dataset and retraining the model, which is unnecessary and resource-intensive when the goal is to follow a single example structure without modifying the underlying model. Option C is wrong because setting the temperature parameter to a high value increases randomness and creativity in the output, which would likely cause the model to deviate from the desired structured format rather than adhere to it. Option D is wrong because the Azure OpenAI embeddings API is used for semantic similarity and search tasks (e.g., finding related content), not for generating structured text outputs like blog outlines.

Full explanation →

248

Matchingmedium

Match each Azure AI service to its use case.

Drag a concept onto its matching description — or click a concept then click the description.

Concepts

Matches

Detect offensive content

Deliver personalized recommendations

Identify unusual patterns in time series

Help users with reading comprehension

Interpret user intents from text

Why these pairings

Each service addresses a specific business problem.

Full explanation →

249

MCQmedium

What is 'scene understanding' in Azure AI Vision?

A.Classifying images by the type of filming location (indoor, outdoor, urban, rural)

B.Holistic comprehension of an image's full context, relationships, and scene description

C.Breaking an image into individual scenes for video timeline analysis

D.Determining the camera settings (ISO, aperture) used to capture a photograph

AnswerB

Scene understanding produces rich contextual descriptions ('red car parked by glass office building') — beyond mere object lists.

Why this answer

Scene understanding in Azure AI Vision goes beyond simple image classification to provide a holistic comprehension of an image's full context, including objects, their relationships, and a descriptive scene summary. This capability leverages deep learning models to analyze the entire visual content and generate human-readable captions that describe what is happening in the image, such as 'a group of people playing soccer in a park.'

Exam trap

The trap here is that candidates often confuse scene understanding with simpler image classification or metadata extraction, leading them to pick options like A or D, which describe narrower tasks rather than the holistic contextual analysis that defines scene understanding.

How to eliminate wrong answers

Option A is wrong because classifying images by filming location (indoor, outdoor, urban, rural) is a specific type of image classification or domain detection, not the comprehensive scene understanding that includes object relationships and full context. Option C is wrong because breaking an image into individual scenes for video timeline analysis is a video analysis task (e.g., shot detection or keyframe extraction), not a core capability of Azure AI Vision's scene understanding feature, which operates on static images. Option D is wrong because determining camera settings like ISO and aperture is metadata extraction or EXIF analysis, which is unrelated to the semantic understanding of image content provided by scene understanding.

Full explanation →

250

MCQmedium

What is 'abstractive summarisation' and how does it differ from 'extractive summarisation'?

A.Extractive writes shorter summaries; abstractive writes longer ones

B.Extractive selects key sentences verbatim; abstractive generates new sentences capturing the meaning

C.Abstractive summarisation is only available for non-English languages

D.Extractive summarisation uses generative AI; abstractive uses keyword ranking

AnswerB

Extractive = copy important sentences; abstractive = write new sentences — abstractive requires deeper language understanding.

Why this answer

Option B is correct because extractive summarisation works by selecting and concatenating the most important sentences directly from the source text, while abstractive summarisation uses natural language generation (NLG) models to produce entirely new sentences that paraphrase and condense the core meaning. This distinction is fundamental in Azure AI Language's summarisation capabilities, where extractive returns verbatim excerpts and abstractive generates novel, coherent summaries.

Exam trap

The trap here is that candidates confuse 'abstractive' with 'longer summaries' or assume it is language-specific, when the core distinction is whether the output uses verbatim sentences (extractive) or generates new sentences (abstractive).

How to eliminate wrong answers

Option A is wrong because the length of the summary is not the defining difference; extractive summaries can be short or long depending on the compression ratio, and abstractive summaries are typically concise but not inherently longer. Option C is wrong because abstractive summarisation is available for multiple languages, including English, and is not restricted to non-English languages. Option D is wrong because extractive summarisation does not use generative AI; it relies on sentence scoring and ranking algorithms, while abstractive summarisation uses generative models (e.g., transformer-based NLG) to create new text.

Full explanation →

251

MCQeasy

What is the difference between Azure AI Speech and Azure AI Language?

A.Azure AI Speech is for transcription only; Azure AI Language handles all other NLP tasks

B.Azure AI Speech handles audio processing; Azure AI Language processes and understands text

C.Azure AI Speech works only in English; Azure AI Language supports multiple languages

D.They are the same service with different pricing tiers

AnswerB

Speech = audio (STT, TTS, speaker recognition); Language = text understanding (sentiment, NER, summarization, classification).

Why this answer

Azure AI Speech is designed to process audio input, converting speech to text (speech-to-text) and text to speech (text-to-speech), as well as enabling speaker recognition and real-time translation. Azure AI Language, on the other hand, processes and understands text by providing capabilities such as sentiment analysis, key phrase extraction, language detection, and question answering. Option B correctly captures this fundamental division: Speech handles audio processing, while Language handles text understanding.

Exam trap

The trap here is that candidates often confuse the scope of Azure AI Speech, mistakenly thinking it only does transcription (Option A), when in fact it also performs text-to-speech and speech translation, while Azure AI Language is strictly for text-based NLP tasks.

How to eliminate wrong answers

Option A is wrong because Azure AI Speech is not limited to transcription; it also includes text-to-speech, speech translation, and speaker recognition, making it more than just a transcription service. Option C is wrong because Azure AI Speech supports multiple languages (e.g., en-US, zh-CN, fr-FR, de-DE) and is not restricted to English, while Azure AI Language also supports a wide range of languages. Option D is wrong because they are distinct services with different APIs, SDKs, and use cases—Speech focuses on audio processing, Language on text analysis—and they are not the same service with different pricing tiers.

Full explanation →

252

MCQmedium

A retail company uses security cameras to monitor shelves. They want to identify whether a customer is holding a specific product (e.g., a green detergent bottle) and also determine the location of that product within the camera frame. Which Azure Computer Vision capability should they use?

A.Object detection

B.Image classification

C.Optical character recognition (OCR)

D.Semantic segmentation

AnswerA

Object detection finds instances of objects and provides bounding box coordinates, which meets both the identification and localization requirements.

Why this answer

Object detection is the correct capability because it not only identifies the presence of a specific product (like a green detergent bottle) in an image but also returns bounding box coordinates that indicate the product's location within the camera frame. This dual output—classification plus localization—directly matches the requirement to both recognize the object and determine its position.

Exam trap

The trap here is that candidates often confuse object detection with image classification, thinking that identifying the product is sufficient, but they overlook the explicit requirement for location information that only object detection provides.

How to eliminate wrong answers

Option B is wrong because image classification assigns a single label to the entire image (e.g., 'detergent bottle') but does not provide any spatial information about where the object is located. Option C is wrong because OCR is designed to extract text from images, not to identify or locate physical products like a detergent bottle. Option D is wrong because semantic segmentation assigns a class label to every pixel in the image, creating a pixel-level mask, but it does not output bounding boxes or directly indicate the product's location within the frame in a way that is typically used for product detection tasks.

Full explanation →

253

MCQmedium

What is 'Azure Machine Learning pipelines' and why are they used?

A.Network pipelines for transferring data between Azure regions at high speed

B.Reusable orchestrated workflows that automate and version-control the full ML training lifecycle

C.CI/CD pipelines in Azure DevOps for deploying application code to production

D.Data pipelines that ingest streaming data from IoT sensors into Azure storage

AnswerB

ML pipelines chain data prep, training, and evaluation steps with caching and scheduling — enabling reproducible, automated ML workflows.

Why this answer

Azure Machine Learning pipelines are reusable orchestrated workflows that automate and version-control the full ML training lifecycle, including data preparation, training, evaluation, and deployment. They enable reproducibility, parallel execution of steps, and easy sharing across teams, which is why option B is correct.

Exam trap

The trap here is that candidates confuse 'pipeline' in the context of ML with generic data or DevOps pipelines, leading them to select options that describe unrelated Azure services like Azure DevOps CI/CD or IoT data ingestion.

How to eliminate wrong answers

Option A is wrong because Azure Machine Learning pipelines are not network pipelines; they are ML-specific workflows, and high-speed data transfer between regions is handled by services like Azure ExpressRoute or Azure Data Box, not ML pipelines. Option C is wrong because CI/CD pipelines in Azure DevOps are for deploying application code, not for orchestrating ML training workflows; ML pipelines focus on the ML lifecycle, not general software deployment. Option D is wrong because data pipelines that ingest streaming data from IoT sensors into Azure storage are typically built with Azure Stream Analytics or Azure IoT Hub, not Azure Machine Learning pipelines, which are designed for ML model training and management.

Full explanation →

254

MCQmedium

A data scientist has a dataset containing information about houses: size (sq ft), number of bedrooms, location, and the actual sale price. The goal is to train a model that predicts the price of a new house based on these features. Which type of machine learning task is this?

A.A) Classification

B.B) Regression

C.C) Clustering

D.D) Reinforcement Learning

AnswerB

Correct. Regression models can predict a continuous numeric output such as house prices.

Why this answer

This is a regression task because the goal is to predict a continuous numeric value (the sale price) based on input features. Regression models learn the relationship between independent variables (size, bedrooms, location) and a dependent variable (price) to output a real number. In Azure Machine Learning, regression algorithms like Linear Regression, Decision Forest Regression, or Neural Network Regression would be appropriate for this scenario.

Exam trap

The trap here is confusing regression with classification because both are supervised learning, but regression outputs a continuous number while classification outputs a discrete label.

How to eliminate wrong answers

Option A is wrong because classification predicts discrete categorical labels (e.g., 'expensive' or 'cheap'), not a continuous numeric price. Option C is wrong because clustering groups unlabeled data into clusters based on similarity, without a target variable like sale price. Option D is wrong because reinforcement learning involves an agent learning through rewards and punishments from interactions with an environment, not from a static dataset with labeled examples.

Full explanation →

255

MCQmedium

What is 'model registry' in Azure Machine Learning?

A.A public marketplace where organisations can buy pre-trained models from third parties

B.A centralised versioned store for tracking and managing trained models and their lineage

C.A database of domain-specific vocabularies used for NLP model training

D.A compliance register documenting AI models used by an organisation for audit purposes

AnswerB

The model registry stores all model versions with metadata and lineage — enabling comparison, rollback, and controlled deployment.

Why this answer

The model registry in Azure Machine Learning is a centralized, versioned store that tracks trained models along with their metadata, lineage, and lifecycle. It enables data scientists to register, version, and manage models, ensuring reproducibility and governance across the ML lifecycle.

Exam trap

The trap here is that candidates confuse the model registry with a marketplace or compliance tool, but the exam specifically tests the registry's role as a versioned repository for managing model artifacts and their lineage.

How to eliminate wrong answers

Option A is wrong because it describes a model marketplace or catalog (like Azure AI Gallery or Hugging Face), not the model registry which is for internal versioning and management. Option C is wrong because it refers to a domain-specific vocabulary database used in NLP, which is unrelated to model tracking; Azure ML uses datasets and tokenizers for such purposes. Option D is wrong because it describes a compliance register for audit purposes, which is a governance artifact, not the model registry's primary function of versioned storage and lineage tracking.

Full explanation →

256

MCQeasy

What is 'computer vision' as a category of AI workload?

A.The display technology used in computer monitors and screens

B.AI capabilities that interpret and understand images, video, and visual information

C.Software for designing user interfaces and graphical layouts

D.A programming paradigm for writing code that processes visual data efficiently

AnswerB

Computer vision gives machines the ability to 'see' — enabling classification, detection, OCR, face analysis, and video understanding.

Why this answer

Computer vision is an AI workload category that enables systems to extract meaningful information from digital images, videos, and other visual inputs. It involves techniques like object detection, image classification, facial recognition, and optical character recognition (OCR), allowing machines to interpret and act on visual data. This is distinct from display hardware or UI design, as it focuses on understanding content rather than rendering or creating it.

Exam trap

The trap here is that candidates confuse 'computer vision' with hardware or software tools for creating visual content, rather than recognizing it as an AI workload that interprets and understands visual information.

How to eliminate wrong answers

Option A is wrong because it describes physical display technology (e.g., LCD, OLED panels), not an AI workload that interprets visual data. Option C is wrong because it refers to software for designing user interfaces and graphical layouts (e.g., Figma, Sketch), which is a design discipline, not an AI capability. Option D is wrong because it misrepresents computer vision as a programming paradigm (like functional or object-oriented programming), whereas it is a category of AI workload that uses specialized algorithms and models (e.g., convolutional neural networks) to process visual information.

Full explanation →

257

MCQeasy

What is 'Azure AI Language's pre-built models' vs 'custom models' and when do you choose each?

A.Pre-built models are free; custom models have additional training costs

B.Pre-built models need no training for general tasks; custom models train on your data for specialised needs

C.Pre-built models only work in English; custom models support all languages

D.Custom models are always more accurate than pre-built regardless of the use case

AnswerB

Pre-built = instant, general purpose. Custom = trained on your labels for domain-specific entity types and categories.

Why this answer

Option B is correct because Azure AI Language provides pre-built models that are ready to use for common NLP tasks like sentiment analysis, key phrase extraction, and language detection without any training. Custom models, on the other hand, require you to upload your own labeled data and train a model to handle specialized needs, such as custom entity recognition or custom text classification, which pre-built models cannot address.

Exam trap

The trap here is that candidates assume pre-built models are free or only support English, when in fact they are paid per use and support many languages, leading them to incorrectly eliminate Option B.

How to eliminate wrong answers

Option A is wrong because pre-built models are not free; they incur consumption-based costs per API call, while custom models also have training costs plus inference costs. Option C is wrong because pre-built models support multiple languages (e.g., sentiment analysis supports over 90 languages), not just English. Option D is wrong because custom models are not always more accurate; they are only better when the pre-built model does not cover your specific domain or terminology, and accuracy depends on the quality and quantity of your training data.

Full explanation →

258

MCQmedium

Which Azure service provides a no-code/low-code drag-and-drop interface for building machine learning pipelines?

A.Azure AI Custom Vision

B.Azure Machine Learning Designer

C.Azure AI Language Studio

D.Azure Databricks

AnswerB

Azure ML Designer provides a visual drag-and-drop interface for building machine learning pipelines without writing code.

Why this answer

Azure Machine Learning Designer is the correct answer because it provides a drag-and-drop, no-code/low-code visual interface for building, testing, and deploying machine learning pipelines. Users can connect pre-built modules for data transformation, model training, and scoring without writing code, making it ideal for rapid prototyping and operationalization of ML workflows.

Exam trap

The trap here is that candidates confuse Azure AI Language Studio (a no-code NLP tool) with a general ML pipeline builder, but Language Studio is domain-specific to text analytics and does not support building arbitrary ML pipelines with drag-and-drop modules.

How to eliminate wrong answers

Option A is wrong because Azure AI Custom Vision is a specialized service for training custom image classification and object detection models, not a general-purpose drag-and-drop ML pipeline builder. Option C is wrong because Azure AI Language Studio is a no-code tool for building natural language processing (NLP) applications like text analysis and conversational AI, not for constructing end-to-end ML pipelines. Option D is wrong because Azure Databricks is a big data analytics and collaborative notebook environment based on Apache Spark, requiring code (Python, Scala, SQL) and lacking a native drag-and-drop pipeline designer.

Full explanation →

259

MCQeasy

What is GitHub Copilot and how does it use AI?

A.An automated GitHub Actions workflow for running CI/CD pipelines

B.An AI-powered code assistant that generates code completions and suggestions in IDEs using LLMs

C.A bot that automatically reviews and merges GitHub pull requests

D.A GitHub feature for visualizing code repository history

AnswerB

GitHub Copilot uses LLMs to suggest code completions, generate functions, explain code, and write tests directly in the development environment.

Why this answer

GitHub Copilot is an AI-powered code assistant developed by GitHub and OpenAI. It uses large language models (LLMs), specifically a version of OpenAI's Codex model, to analyze the context of the code a developer is writing in an IDE (like VS Code) and generate real-time code completions, suggestions, and even entire functions. This directly aligns with generative AI workloads on Azure, as Copilot leverages generative AI to produce new code content based on natural language prompts or existing code patterns.

Exam trap

The trap here is that candidates confuse GitHub Copilot with GitHub Actions or other automation features, because all are GitHub services, but Copilot is specifically a generative AI code assistant, not a CI/CD or repository management tool.

How to eliminate wrong answers

Option A is wrong because GitHub Copilot is not an automated CI/CD workflow; GitHub Actions is the service that runs CI/CD pipelines, and Copilot is a code generation tool, not a workflow executor. Option C is wrong because Copilot does not automatically review or merge pull requests; that is the function of tools like GitHub's built-in pull request review features or third-party bots (e.g., Dependabot). Option D is wrong because Copilot does not visualize repository history; that is handled by GitHub's Insights or git log commands, not by an AI code assistant.

Full explanation →

260

MCQhard

A company uses Azure OpenAI Service to generate marketing copy for social media posts. They want to prevent the model from producing content that contains offensive language, harmful stereotypes, or violent themes that go against their brand guidelines. Which feature should the company configure within Azure OpenAI Service?

A.Fine-tuning the model with a custom dataset

B.Configuring the content filtering (responsible AI filters)

C.Increasing the token limit per response

D.Using prompt engineering techniques

AnswerB

Azure OpenAI’s content filtering system is a built-in safeguard that automatically screens inputs and outputs for categories like hate, violence, sexual content, and self-harm. Companies can configure severity levels to prevent undesirable content from being generated.

Why this answer

B is correct because Azure OpenAI Service includes built-in content filtering (responsible AI filters) that automatically detects and blocks offensive language, harmful stereotypes, and violent themes in both input prompts and generated outputs. This feature enforces brand guidelines without requiring custom model modifications or manual oversight.

Exam trap

The trap here is that candidates often confuse fine-tuning or prompt engineering as content safety mechanisms, when in fact Azure OpenAI's content filtering is the only built-in feature designed specifically to block offensive or harmful content at inference time.

How to eliminate wrong answers

Option A is wrong because fine-tuning adapts the model to specific tasks or styles using custom data but does not enforce safety guardrails; it can even amplify harmful patterns if the training data contains them. Option C is wrong because increasing the token limit per response only controls the maximum length of generated text, not its content safety. Option D is wrong because prompt engineering techniques can guide model behavior but are not a reliable or enforceable mechanism to prevent offensive content—they can be bypassed by adversarial inputs.

Full explanation →

261

MCQmedium

A retail company uses overhead cameras to monitor shelf inventory in a store. They want to build a system that automatically detects whether a shelf section is empty or stocked, and specifically identify product categories (e.g., 'soft drinks', 'chips', 'canned goods') and count the number of items in each category. The company has a large set of labeled images showing different shelf states. Which Azure Computer Vision service should they use to build this custom detection and counting solution?

A.Computer Vision Image Analysis with dense captioning

B.Custom Vision object detection

C.Optical Character Recognition (OCR)

D.Azure Machine Learning with a pre-trained YOLO model

AnswerB

Custom Vision object detection is specifically designed for training models to detect and locate objects of interest. With labeled images of product categories, you can create a model that outputs bounding boxes around each detected item, enabling counting.

Why this answer

Custom Vision object detection is the correct choice because it allows the company to train a model on their labeled images to detect and localize specific product categories (e.g., 'soft drinks', 'chips') and count items within each category. Unlike pre-built Computer Vision features, Custom Vision enables custom object detection with bounding boxes and classification, which directly supports the requirement for detecting shelf states and counting items per category.

Exam trap

The trap here is that candidates confuse pre-built Computer Vision features (like dense captioning or OCR) with Custom Vision, assuming any Azure Computer Vision service can be customized without training, but only Custom Vision supports custom object detection with bounding boxes and counting.

How to eliminate wrong answers

Option A is wrong because Computer Vision Image Analysis with dense captioning generates descriptive captions for regions of an image but does not provide object detection with bounding boxes or item counting per category, making it unsuitable for precise inventory counting. Option C is wrong because Optical Character Recognition (OCR) extracts text from images, which is irrelevant for detecting non-textual objects like product categories on shelves. Option D is wrong because Azure Machine Learning with a pre-trained YOLO model, while technically capable, is not a managed Azure Computer Vision service; the question asks for an Azure Computer Vision service, and Custom Vision is the appropriate PaaS offering for custom object detection without requiring manual ML pipeline setup.

Full explanation →

262

MCQmedium

What does Azure AI Vision's 'people detection' (spatial analysis) feature track?

A.Identifying the names of specific people in video footage

B.Counting, tracking movement, and measuring occupancy of people in defined zones from video

C.Detecting whether people are wearing masks or safety equipment

D.Measuring individual people's heights and body dimensions

AnswerB

Spatial analysis tracks anonymous people to measure occupancy, queue length, zone entry/exit, and dwell time.

Why this answer

Azure AI Vision's spatial analysis (people detection) tracks the movement of people in video feeds, counting individuals and measuring how long they stay in defined zones. It does not identify specific people, detect masks or safety equipment, or measure body dimensions. This feature is designed for occupancy monitoring and flow analysis in physical spaces.

Exam trap

The trap here is that candidates confuse 'people detection' with facial recognition or attribute detection (like masks), but Azure AI Vision's spatial analysis is strictly about anonymous tracking and counting, not identification or detailed attribute analysis.

How to eliminate wrong answers

Option A is wrong because Azure AI Vision's people detection does not perform facial recognition or identify specific individuals; it only detects and tracks people as anonymous objects. Option C is wrong because detecting masks or safety equipment is a separate custom vision capability, not part of the spatial analysis people detection feature. Option D is wrong because the feature does not measure individual heights or body dimensions; it only tracks presence, movement, and occupancy in zones.

Full explanation →

263

MCQmedium

A data scientist trains a classification model to predict whether an email is spam or not. The model achieves 98% accuracy on the test set, but upon inspection, it classifies all emails as 'not spam' because the dataset has 95% non-spam emails. What is the most likely issue?

A.Overfitting

B.Underfitting

C.Data imbalance

D.Feature scaling error

AnswerC

Data imbalance, where one class vastly outnumbers the other, can cause a model to predict the majority class exclusively. Accuracy is misleading in such cases; the model has not learned to identify spam.

Why this answer

The model achieves 98% accuracy by simply predicting all emails as 'not spam', which reflects the 95% majority class in the dataset. This is a classic symptom of class imbalance, where the model learns to exploit the skewed distribution rather than learning meaningful patterns to distinguish spam from non-spam. In Azure Machine Learning, techniques like SMOTE or stratified sampling are used to mitigate this issue.

Exam trap

The trap here is that candidates see 98% accuracy and assume the model is performing well, failing to recognize that accuracy is meaningless when the dataset is highly imbalanced and the model simply predicts the majority class.

How to eliminate wrong answers

Option A is wrong because overfitting would cause the model to perform well on training data but poorly on unseen test data, whereas here the model performs uniformly poorly on the minority class across both sets. Option B is wrong because underfitting would result in low accuracy on both training and test sets due to insufficient model complexity, not high accuracy driven by majority class bias. Option D is wrong because feature scaling errors affect models sensitive to input ranges (e.g., SVM, neural networks), but the issue here is purely about class distribution, not feature preprocessing.

Full explanation →

264

MCQmedium

A customer support team wants to automatically analyze thousands of product reviews. Their goal is to extract the most frequently mentioned topics (e.g., 'battery life', 'customer service', 'screen quality') without manually reading each review. Which Azure AI Language feature should they use?

A.Key phrase extraction

B.Sentiment analysis

C.Language detection

D.Entity recognition

AnswerA

Key phrase extraction is designed to identify the main concepts and topics in a body of text, making it ideal for extracting frequently mentioned subjects from product reviews.

Why this answer

Key phrase extraction is the correct Azure AI Language feature because it automatically identifies the main points or topics (e.g., 'battery life', 'customer service', 'screen quality') from unstructured text. This directly meets the requirement to extract frequently mentioned topics from thousands of product reviews without manual reading.

Exam trap

The trap here is that candidates often confuse 'entity recognition' with 'key phrase extraction', but entity recognition only extracts proper nouns (e.g., 'Apple', 'New York'), while key phrase extraction captures descriptive multi-word topics (e.g., 'battery life').

How to eliminate wrong answers

Option B (Sentiment analysis) is wrong because it determines the overall positive, negative, or neutral emotional tone of text, not the specific topics or themes mentioned. Option C (Language detection) is wrong because it identifies the language of the text (e.g., English, Spanish), not the subject matter or key topics. Option D (Entity recognition) is wrong because it identifies named entities such as people, organizations, or locations, not general descriptive phrases like 'battery life' or 'screen quality'.

Full explanation →

265

MCQhard

What is 'chain-of-thought prompting' and when is it most effective?

A.Linking multiple AI models in a pipeline where each model's output feeds the next

B.Prompting the model to show explicit reasoning steps before giving a final answer

C.Training a model on a sequence of related documents to build contextual knowledge

D.A method for connecting chatbot conversation turns to maintain long-term memory

AnswerB

CoT prompting ('think step by step') improves multi-step reasoning by externalising the reasoning process — most effective for maths and logic.

Why this answer

Chain-of-thought prompting instructs the model to break down a complex problem into intermediate reasoning steps before producing the final answer. This technique improves accuracy on tasks requiring multi-step logic, such as arithmetic, commonsense reasoning, or symbolic manipulation, by making the model's internal reasoning explicit and reducing errors from shortcut answers.

Exam trap

The trap here is that candidates confuse 'chain-of-thought prompting' with 'model chaining' or 'pipeline architectures' (Option A), because both involve a sequence, but chain-of-thought is a single-model prompting technique, not a multi-model workflow.

How to eliminate wrong answers

Option A is wrong because it describes a model pipeline or ensemble, not a prompting technique; chain-of-thought prompting does not involve linking multiple models. Option C is wrong because it describes sequential training or fine-tuning on related documents, which is a data preparation or transfer learning approach, not a prompting strategy. Option D is wrong because it describes conversation memory or state management in chatbots, which is unrelated to the explicit step-by-step reasoning elicited by chain-of-thought prompts.

Full explanation →

266

MCQhard

A data scientist is training a credit risk model and wants to use Azure Machine Learning's Responsible AI dashboard to identify if the model is biased against a certain demographic group. Which component of the dashboard should they use to evaluate this?

A.Model Interpretability

B.Model Fairness Assessment

C.Error Analysis

D.Data Balance Analysis

AnswerB

This component analyzes model predictions across predefined sensitive groups to identify and measure unfair bias.

Why this answer

The Model Fairness Assessment component of Azure Machine Learning's Responsible AI dashboard is specifically designed to evaluate and mitigate bias in machine learning models. It allows data scientists to assess disparities in model performance across demographic groups defined by sensitive features (e.g., race, gender) using metrics like demographic parity, equal opportunity, and disparate impact. This directly addresses the question of identifying bias against a certain demographic group.

Exam trap

The trap here is that candidates often confuse Model Interpretability (which explains why a model made a prediction) with Fairness Assessment (which evaluates bias across groups), leading them to select Option A when the question specifically asks about bias against a demographic group.

How to eliminate wrong answers

Option A is wrong because Model Interpretability focuses on explaining model predictions (e.g., feature importance, SHAP values) rather than measuring bias or fairness across demographic groups. Option C is wrong because Error Analysis is used to identify and understand error patterns in model predictions (e.g., error distribution by feature values), not to evaluate fairness or bias against protected groups. Option D is wrong because Data Balance Analysis assesses the distribution and representation of data features (e.g., class imbalance, missing values), but does not directly evaluate model bias or fairness metrics across demographic groups.

Full explanation →

267

MCQmedium

What is Azure AI Document Intelligence's 'custom extraction model' used for?

A.Automatically generating new document templates from existing forms

B.Training on your labeled documents to extract business-specific fields not covered by prebuilt models

C.Translating documents into multiple languages simultaneously

D.Redacting sensitive information from documents automatically

AnswerB

Custom extraction models handle unique business forms — labeled with your specific field names to train field extraction for your document types.

Why this answer

Azure AI Document Intelligence's custom extraction model is correct because it allows you to train a model on your own labeled documents to extract fields that are specific to your business domain and not covered by prebuilt models. This is essential for processing specialized forms like invoices, contracts, or medical records that have unique data fields.

Exam trap

The trap here is that candidates often confuse custom extraction models with template generation or translation, assuming Document Intelligence can create templates or translate text, when in reality it is strictly for extraction and classification of document content.

How to eliminate wrong answers

Option A is wrong because custom extraction models do not generate new document templates; they learn to extract specific fields from existing documents, not create templates. Option C is wrong because document translation is handled by Azure AI Translator, not Document Intelligence, which focuses on extraction and classification. Option D is wrong because redaction of sensitive information is not a built-in feature of custom extraction models; it would require additional processing or integration with other services like Azure Purview or custom logic.

Full explanation →

268

MCQmedium

What does the responsible AI principle of 'human in the loop' refer to?

A.A requirement for humans to manually enter all data into AI systems

B.Maintaining human oversight and the ability to review or override consequential AI decisions

C.Training AI models using feedback from human labelers only

D.Requiring users to prove they are human before using AI services

AnswerB

'Human in the loop' ensures humans remain in control of high-stakes AI decisions, with the ability to review, override, or veto AI outputs.

Why this answer

The 'human in the loop' principle ensures that humans maintain meaningful oversight over AI systems, particularly for high-stakes or consequential decisions. This means humans can review, override, or intervene in AI-generated outputs, preventing fully automated decision-making in critical scenarios such as medical diagnosis, loan approvals, or criminal justice. It is a core component of responsible AI, balancing automation with accountability.

Exam trap

The trap here is confusing 'human in the loop' with general human involvement (like data entry or CAPTCHA) rather than recognizing it specifically as oversight of consequential AI decisions.

How to eliminate wrong answers

Option A is wrong because 'human in the loop' does not require manual data entry; it focuses on oversight of decisions, not data ingestion. Option C is wrong because while human labelers may be used in training, the principle is about ongoing human review of AI outputs, not exclusively about training data sources. Option D is wrong because CAPTCHA-style human verification is a security measure, not a responsible AI principle for oversight of consequential decisions.

Full explanation →

269

MCQeasy

A healthcare company develops an AI system to recommend treatment plans. The system sometimes provides recommendations that contradict standard medical guidelines, leading to potential patient harm. Which Microsoft responsible AI principle is most directly violated?

A.Fairness

B.Reliability and safety

C.Privacy and security

D.Inclusiveness

AnswerB

This principle focuses on ensuring AI systems perform consistently and safely, which is directly challenged by a system that gives harmful, incorrect recommendations.

Why this answer

The system's recommendations contradicting standard medical guidelines and causing potential patient harm directly violates the Reliability and safety principle. This principle requires AI systems to perform consistently, safely, and as intended, especially in high-stakes domains like healthcare where failures can lead to injury or death. The scenario describes a lack of robustness and failure to meet expected safety standards, which is the core concern of this principle.

Exam trap

The trap here is that candidates may confuse 'safety' with 'fairness' or 'privacy,' but the key indicator is the direct mention of 'patient harm' and 'contradicting standard medical guidelines,' which points squarely to the Reliability and safety principle.

How to eliminate wrong answers

Option A is wrong because Fairness focuses on ensuring AI systems do not discriminate against groups or individuals based on attributes like race or gender; the issue here is about safety and correctness of recommendations, not bias. Option C is wrong because Privacy and security concern the protection of personal data and system integrity from unauthorized access or breaches; the problem is about the system's output contradicting medical guidelines, not data exposure. Option D is wrong because Inclusiveness aims to empower everyone and design for diverse user needs, including accessibility; the scenario does not describe exclusion or lack of accessibility, but rather unsafe recommendations.

Full explanation →

270

MCQmedium

What is the 'presence penalty' parameter in Azure OpenAI API calls?

A.A parameter requiring AI systems to acknowledge their presence as AI to users

B.A flat penalty discouraging repetition of any token already present in the response

C.A parameter indicating whether the AI is present online or offline

D.The minimum number of characters that must be present in a response

AnswerB

Presence penalty adds a flat penalty for any previously used token — encouraging vocabulary diversity regardless of repeat frequency.

Why this answer

The 'presence penalty' parameter in Azure OpenAI API calls applies a flat penalty to any token that has already appeared in the response so far, reducing the model's likelihood of repeating that token. This helps generate more diverse and less repetitive text by discouraging the reuse of tokens already present in the output sequence.

Exam trap

Microsoft often tests the distinction between 'presence penalty' and 'frequency penalty' — the trap here is that candidates confuse the presence penalty with a requirement for AI disclosure or a simple repetition penalty, missing that it specifically penalizes any token that has already appeared at least once, regardless of how many times.

How to eliminate wrong answers

Option A is wrong because it describes a transparency or disclosure requirement (like an AI disclosure policy), not a parameter that modifies token probabilities in the API. Option C is wrong because it confuses a presence/availability status with a model inference parameter; Azure OpenAI does not have an 'online/offline' parameter in API calls. Option D is wrong because it describes a minimum length constraint, which is unrelated to the presence penalty; the presence penalty operates on token-level repetition, not character count.

Full explanation →

271

MCQeasy

What is 'the AI-900 exam' testing you on?

A.Advanced ML model development and Azure ML pipeline coding skills

B.Foundational knowledge of AI/ML concepts and Azure AI services — suitable for non-technical stakeholders

C.Azure infrastructure management and deployment of AI workloads using IaC tools

D.Ethical AI policy writing and regulatory compliance documentation

AnswerB

AI-900 covers AI fundamentals and Azure AI service capabilities — no coding required, aimed at business roles and AI newcomers.

Why this answer

The AI-900 exam is designed to validate foundational knowledge of AI and machine learning concepts, along with familiarity with Azure AI services. It targets non-technical stakeholders, such as business analysts or project managers, who need to understand AI capabilities and ethical considerations without requiring hands-on coding or infrastructure skills.

Exam trap

The trap here is that candidates often confuse AI-900 with a technical implementation exam, assuming it requires coding or infrastructure skills, when it actually tests conceptual understanding suitable for non-technical roles.

How to eliminate wrong answers

Option A is wrong because it describes advanced ML model development and Azure ML pipeline coding, which are topics for the AI-102 or DP-100 exams, not the foundational AI-900. Option C is wrong because Azure infrastructure management and deployment using IaC tools (e.g., ARM templates, Terraform) are covered in Azure Administrator (AZ-104) or DevOps exams, not AI-900. Option D is wrong because ethical AI policy writing and regulatory compliance documentation are not the primary focus; AI-900 covers ethical AI principles at a conceptual level, not policy creation or compliance documentation.

Full explanation →

272

Drag & Dropmedium

Drag and drop the steps to analyze an image with Azure Computer Vision into the correct order.

Drag steps to the numbered slots on the right, or tap a step then tap a slot.

Steps

Order

Why this order

Image analysis requires resource setup, API call with features, and understanding the response.

Full explanation →

273

MCQmedium

A financial institution uses an AI model to approve loan applications. A customer is denied a loan and requests an explanation for the decision. The development team cannot explain how the model reached its conclusion because the model is a deep neural network with complex layers. Which Microsoft responsible AI principle is being violated in this scenario?

A.Fairness

B.Accountability

C.Reliability and Safety

D.Privacy and Security

AnswerB

Accountability requires that people can understand and explain AI decisions. The team cannot explain the decision, violating this principle.

Why this answer

The scenario describes a model whose internal reasoning cannot be explained by the development team, which directly violates the Microsoft responsible AI principle of Accountability. Accountability requires that organizations be able to explain AI decisions and take ownership of their outcomes. A deep neural network that acts as a 'black box' prevents the team from providing the required explanation to the customer, thus breaking this principle.

Exam trap

Microsoft often tests the distinction between Accountability and Fairness, where candidates mistakenly choose Fairness because they associate 'explanation' with 'bias detection,' but the core issue here is the inability to explain the decision, not the presence of bias.

How to eliminate wrong answers

Option A is wrong because Fairness deals with bias and equitable treatment across groups, not with the ability to explain a specific decision. Option C is wrong because Reliability and Safety focuses on whether the model performs consistently and safely under expected conditions, not on post-hoc explainability. Option D is wrong because Privacy and Security concerns data protection and unauthorized access, not the interpretability of model outputs.

Full explanation →

274

MCQmedium

A data scientist trains a classification model on a dataset of 10,000 labeled emails to distinguish spam from non-spam. The model achieves 99% accuracy on the training data but only 70% accuracy on a held-out test set. Which term best describes this situation?

A.A) Underfitting

B.B) Overfitting

C.C) Bias-variance tradeoff

D.D) Regularization

AnswerB

Overfitting happens when the model memorizes the training data, including noise, leading to high training accuracy but low test accuracy. This matches the described 99% training vs 70% test accuracy.

Why this answer

The model performs exceptionally well on training data (99% accuracy) but poorly on unseen test data (70% accuracy), which is the classic symptom of overfitting. Overfitting occurs when the model learns noise and specific patterns in the training set rather than generalizing to new data, often due to excessive model complexity or insufficient regularization.

Exam trap

The trap here is that candidates confuse 'high training accuracy' with a good model, failing to recognize that the large gap between training and test performance is the hallmark of overfitting, not underfitting or a tradeoff concept.

How to eliminate wrong answers

Option A is wrong because underfitting would result in poor performance on both training and test data, not high training accuracy with low test accuracy. Option C is wrong because bias-variance tradeoff is a general principle describing the balance between underfitting (high bias) and overfitting (high variance), but it is not the specific term for this situation where the model has memorized the training data. Option D is wrong because regularization is a technique used to prevent overfitting (e.g., L1/L2 regularization in Azure ML), not the name of the problem itself.

Full explanation →

275

MCQeasy

What is 'speech recognition' as an AI workload?

A.Identifying which employee is speaking during a meeting using their voice

B.Converting spoken audio into written text

C.Recognising specific wake words to activate voice assistant devices

D.Detecting background noise in audio to improve recording quality

AnswerB

Speech recognition (speech-to-text) transcribes spoken words into text — enabling voice interfaces, transcription, and accessibility.

Why this answer

Speech recognition, also known as automatic speech recognition (ASR), is an AI workload that converts spoken language into written text. It processes audio input and maps it to words using acoustic and language models, enabling transcription, voice commands, and dictation. Option B correctly identifies this core function.

Exam trap

The trap here is that candidates confuse speech recognition with related but distinct tasks like speaker identification (Option A) or wake-word detection (Option C), leading them to pick a narrower or incorrect definition.

How to eliminate wrong answers

Option A is wrong because identifying which employee is speaking based on their voice is speaker recognition (or speaker diarization), not speech recognition; speech recognition focuses on what is said, not who said it. Option C is wrong because recognizing specific wake words (e.g., 'Hey Siri') is a keyword spotting or wake-word detection task, which is a subset of speech recognition but not the full workload definition. Option D is wrong because detecting background noise to improve recording quality is audio enhancement or noise reduction, not speech recognition; speech recognition does not inherently optimize audio quality.

Full explanation →

276

MCQmedium

What is 'AI governance' and what tools does Azure provide for it?

A.Government regulation that prohibits certain types of AI systems

B.The policies, processes, and controls ensuring AI systems are developed and operated responsibly

C.Electing a board of AI experts to approve all AI projects before they go to production

D.Restricting AI development to organisations with formal AI certifications

AnswerB

AI governance covers policies, auditing, fairness monitoring, and compliance tools — Azure ML's Responsible AI dashboard supports this.

Why this answer

AI governance refers to the framework of policies, processes, and controls that guide the responsible development, deployment, and operation of AI systems. Azure provides tools like Azure Policy, Azure Role-Based Access Control (RBAC), and Microsoft Purview to enforce governance rules, audit AI usage, and ensure compliance with ethical standards. Option B correctly captures this definition, as it focuses on the organizational and technical mechanisms for responsible AI, not external restrictions or certifications.

Exam trap

The trap here is that candidates confuse 'AI governance' with external regulation (Option A) or a specific approval process (Option C), rather than recognizing it as the internal framework of policies and controls that Azure implements through tools like Azure Policy and RBAC.

How to eliminate wrong answers

Option A is wrong because it describes government regulation, which is a subset of external legal requirements, not the internal policies and controls that constitute AI governance; Azure's governance tools are about organizational enforcement, not just compliance with prohibitions. Option C is wrong because it suggests a board approval process, which is a specific governance practice but not the definition of AI governance itself; Azure does not mandate such boards, and governance is broader than project approval workflows. Option D is wrong because it implies restricting AI development to certified organizations, which is not a core aspect of AI governance; Azure's governance tools focus on operational controls (e.g., RBAC, policy assignments) rather than external certification requirements.

Full explanation →

277

MCQeasy

What is 'Azure AI Vision's Read API' and what makes it superior for OCR?

A.The standard API for reading data from Azure Storage accounts and databases

B.An advanced OCR service handling multi-page PDFs, handwriting, and complex layouts with word-level coordinates

C.An API for reading audio content and converting it to text transcripts

D.A feature for reading the metadata of image files stored in Azure Blob Storage

AnswerB

The Read API goes beyond basic OCR — multi-page, handwriting, complex layouts, and word positions — powered by deep learning.

Why this answer

Azure AI Vision's Read API is an advanced OCR service that extracts text from images and documents, including multi-page PDFs, handwritten text, and complex layouts. It is superior because it returns word-level bounding box coordinates and confidence scores, enabling precise text localization and structured output for downstream processing.

Exam trap

The trap here is that candidates may confuse the Read API with other Azure services like Storage APIs or Speech services, overlooking that it is specifically a computer vision OCR service for text extraction from images and documents.

How to eliminate wrong answers

Option A is wrong because the Read API is not for reading data from Azure Storage accounts or databases; it is an OCR service for extracting text from visual content. Option C is wrong because reading audio content and converting it to text is the function of Azure Speech-to-Text, not the Read API. Option D is wrong because reading metadata of image files is not the purpose of the Read API; it extracts text content from images, not file metadata.

Full explanation →

278

MCQhard

A global customer support team receives feedback messages in multiple languages. They want to build an automated pipeline that first identifies the language of each message, then translates it to English, and finally analyzes the sentiment of the translated text. Which combination of Azure AI services should they use?

A.Azure AI Translator and Azure Anomaly Detector

B.Azure AI Translator and Azure AI Language (Text Analytics)

C.Azure AI Speech and Azure AI Language (Text Analytics)

D.Azure AI Vision and Azure AI Language (Text Analytics)

AnswerB

Translator detects language and translates; Azure AI Language provides language detection (if needed separately) and sentiment analysis on the translated text.

Why this answer

Option B is correct because the pipeline requires language detection, translation, and sentiment analysis. Azure AI Translator provides language detection and translation, while Azure AI Language (Text Analytics) provides sentiment analysis on the translated English text. This combination directly fulfills all three requirements.

Exam trap

The trap here is that candidates may confuse Azure AI Language (Text Analytics) with Azure AI Speech or Azure AI Vision, mistakenly thinking speech or vision services can handle text-based language detection and sentiment analysis, when in fact only Text Analytics provides those NLP capabilities.

How to eliminate wrong answers

Option A is wrong because Azure Anomaly Detector is used for identifying anomalies in time-series data, not for sentiment analysis or language processing. Option C is wrong because Azure AI Speech handles speech-to-text and text-to-speech, not text translation or sentiment analysis of written feedback messages. Option D is wrong because Azure AI Vision is for image and video analysis, not for language detection, translation, or sentiment analysis.

Full explanation →

279

MCQeasy

A real estate company has a dataset containing square footage, number of bedrooms, and location for 10,000 houses, along with their sale prices. They want to train a model that predicts the sale price of a new house based on these features. Which type of machine learning should they use?

A.Supervised classification

B.Supervised regression

C.Unsupervised clustering

D.Reinforcement learning

AnswerB

Regression is a type of supervised learning that models the relationship between input features and a continuous numeric target variable, such as sale price.

Why this answer

The goal is to predict a continuous numeric value (sale price) from input features (square footage, bedrooms, location). This is a classic supervised regression problem because the training data includes labeled target values (prices) and the output is a real number, not a category.

Exam trap

The trap here is that candidates confuse 'classification' with any prediction task, forgetting that regression is specifically for continuous numeric outputs, not categorical labels.

How to eliminate wrong answers

Option A is wrong because supervised classification predicts discrete class labels (e.g., 'expensive' or 'cheap'), not a continuous numeric price. Option C is wrong because unsupervised clustering finds hidden patterns or groups in unlabeled data, but here the dataset has labeled sale prices, so clustering is inappropriate. Option D is wrong because reinforcement learning learns optimal actions through trial-and-error interactions with an environment (e.g., game playing or robotics), not from a static dataset of house features and prices.

Full explanation →

280

MCQmedium

A customer support team wants to use Azure AI Language to automatically analyze incoming support emails. They need to extract the product name mentioned in each email and determine whether the customer's sentiment is positive, negative, or neutral. They have no labeled data for custom training. Which two prebuilt Azure AI Language features should they use together?

A.Key phrase extraction and language detection

B.Named entity recognition (NER) and sentiment analysis

C.Conversational language understanding (CLU) and translation

D.Text summarization and personal identifying information (PII) detection

AnswerB

NER extracts product names as entities, and sentiment analysis classifies the tone of the email as positive, negative, or neutral. Both are prebuilt and require no custom training.

Why this answer

Named entity recognition (NER) extracts specific entities like product names from text, while sentiment analysis determines the emotional tone (positive, negative, neutral). Both are prebuilt, no-code features in Azure AI Language that require no labeled data for custom training, making them the correct pair for this use case.

Exam trap

Microsoft often tests the distinction between prebuilt features (NER, sentiment analysis) that require no training data versus custom features (CLU) that need labeled data, causing candidates to mistakenly choose CLU for entity extraction without realizing it requires custom training.

How to eliminate wrong answers

Option A is wrong because key phrase extraction returns general keywords (e.g., 'customer service') but not structured entities like product names, and language detection only identifies the language, not sentiment. Option C is wrong because conversational language understanding (CLU) requires labeled training data for custom intent/entity extraction, and translation changes the language but does not extract product names or sentiment. Option D is wrong because text summarization condenses text without extracting specific entities or sentiment, and PII detection identifies sensitive data (e.g., SSNs) but not product names or sentiment.

Full explanation →

281

MCQmedium

A legal firm needs to automatically extract the names of organizations and monetary values from thousands of legal contracts. They want to use a prebuilt Azure AI Language feature without custom training. Which feature should they use?

A.Key phrase extraction

B.Named entity recognition

C.Sentiment analysis

D.Text summarization

AnswerB

Correct. NER is designed to extract and classify entities into predefined categories, including Organization and Money, making it ideal for this task.

Why this answer

Named entity recognition (NER) is the correct choice because it is specifically designed to identify and categorize entities such as organization names and monetary values from unstructured text. Azure AI Language's prebuilt NER model can extract these entity types without any custom training, making it ideal for processing legal contracts at scale.

Exam trap

The trap here is that candidates often confuse key phrase extraction with named entity recognition, assuming that extracting 'important phrases' is equivalent to identifying specific entity types like organizations and monetary values.

How to eliminate wrong answers

Option A is wrong because key phrase extraction identifies important words or phrases (e.g., 'contract terms', 'payment schedule') but does not categorize them into predefined types like organizations or monetary values. Option C is wrong because sentiment analysis evaluates the emotional tone (positive, negative, neutral) of text, not the extraction of named entities. Option D is wrong because text summarization generates a condensed version of the document, not the extraction of specific entity types.

Full explanation →

282

MCQeasy

A hotel booking website wants to automatically analyze guest-submitted photos of hotel rooms to verify if they contain common amenities such as a bed, a desk, and a chair. They want to use a prebuilt Azure AI service without any custom training. Which feature should they use?

A.Optical Character Recognition (OCR)

B.Image Analysis (prebuilt)

C.Object Detection

D.Handwriting OCR

AnswerC

Azure Computer Vision's prebuilt object detection identifies common objects (such as bed, desk, chair) in an image and returns their locations with bounding boxes. This is the correct capability for verifying the presence of specific furniture items.

Why this answer

Object Detection (prebuilt) is the correct choice because it can identify and locate multiple specific objects (bed, desk, chair) within an image by drawing bounding boxes around them. This prebuilt Azure AI Vision feature requires no custom training and directly supports detecting common amenities in hotel room photos.

Exam trap

The trap here is that candidates confuse 'Image Analysis' (which provides descriptive tags but not precise object localization) with 'Object Detection' (which provides bounding boxes for specific objects), leading them to choose Option B incorrectly.

How to eliminate wrong answers

Option A is wrong because Optical Character Recognition (OCR) extracts text from images, not objects like furniture. Option B is wrong because prebuilt Image Analysis provides general tags and descriptions but cannot precisely locate or verify multiple specific objects (e.g., bed, desk, chair) with bounding boxes. Option D is wrong because Handwriting OCR is a specialized form of OCR for handwritten text, not for detecting physical objects.

Full explanation →

283

MCQmedium

A healthcare organization uses an AI system to predict patient readmission risk. The model was trained on data from a single hospital with a predominantly elderly population. When deployed to a different hospital with a younger demographic, the model's accuracy drops significantly. Which responsible AI principle is most directly violated?

A.Transparency

B.Fairness

C.Reliability and safety

D.Privacy and security

AnswerC

This principle requires AI systems to perform as intended across a range of conditions and to be robust to changes in data distribution. The model's failure to generalize to a different demographic violates this principle.

Why this answer

The model's accuracy drop when applied to a different demographic indicates a failure in reliability and safety. The model was trained on a non-representative dataset (elderly patients) and does not generalize to younger populations, violating the principle that AI systems must perform consistently and safely across intended deployment contexts.

Exam trap

The trap here is confusing a model's failure to generalize (reliability) with fairness, because candidates may incorrectly assume that any performance disparity across demographic groups automatically constitutes a fairness violation.

How to eliminate wrong answers

Option A is wrong because transparency refers to the ability to understand and explain how an AI model makes decisions, not to performance degradation across different data distributions. Option B is wrong because fairness concerns bias that leads to discriminatory outcomes against protected groups; while the model may be less accurate for younger patients, this is a generalization failure, not a systematic bias against a protected attribute. Option D is wrong because privacy and security involve protecting data from unauthorized access or misuse, which is not implicated by the model's poor performance on new data.

Full explanation →

284

MCQhard

A medical research organization uses an AI system to analyze patient health records to identify patterns in disease progression. They publish a research paper that includes tables of aggregated statistics derived from the data. Later, a researcher discovers that by combining multiple statistics, it is possible to identify individual patients. Which Microsoft responsible AI principle has been most directly compromised?

A.Fairness

B.Privacy and security

C.Transparency

D.Accountability

AnswerB

This principle mandates protecting personal data and preventing unauthorized identification, which was compromised in this scenario.

Why this answer

The correct answer is B because the scenario describes a re-identification attack, where aggregated statistics (tables) can be combined to infer individual patient identities. This directly violates the privacy and security principle, which requires that AI systems protect personal data and prevent unauthorized identification. Microsoft's responsible AI principle of privacy and security emphasizes safeguarding data through techniques like differential privacy, which was not applied here.

Exam trap

The trap here is that candidates often confuse aggregated statistics with anonymized data, assuming that tables of averages or counts cannot reveal individuals, but re-identification attacks (e.g., via differencing or linking multiple tables) directly compromise privacy and security.

How to eliminate wrong answers

Option A is wrong because fairness is about ensuring AI systems do not discriminate against groups or individuals based on attributes like race or gender; the issue here is data leakage, not bias. Option C is wrong because transparency refers to making AI systems understandable and their decisions explainable; the problem is not a lack of explanation but a failure to protect individual privacy. Option D is wrong because accountability involves assigning responsibility for AI system outcomes and ensuring governance; while a breach occurred, the core compromised principle is privacy and security, not a lack of oversight or ownership.

Full explanation →

285

MCQhard

A bank deploys an AI system that uses a deep neural network to approve personal loan applications. A customer whose loan was rejected requests a detailed explanation of why the decision was made. The bank's AI team realizes that the model's internal workings are too complex to provide a simple, understandable reason. According to Microsoft's responsible AI principles, which principle is most directly violated by this situation?

A.Fairness

B.Transparency

C.Reliability & Safety

D.Privacy & Security

AnswerB

Transparency ensures that AI systems are understandable and that decisions can be explained to users, which is directly missing in this case.

Why this answer

The bank's inability to provide a clear, understandable explanation for the AI's loan decision directly violates the transparency principle. Microsoft's responsible AI principles require that AI systems be understandable and that their decisions can be explained to users, especially when those decisions have significant impact. A deep neural network's complex, non-linear decision boundaries and lack of inherent interpretability make it a 'black box,' which undermines the required transparency.

Exam trap

The trap here is that candidates may confuse 'transparency' with 'fairness,' assuming that an unexplained decision must be biased, but the question specifically tests the principle of providing understandable explanations, not the presence of discrimination.

How to eliminate wrong answers

Option A is wrong because fairness concerns bias or discrimination in outcomes, not the lack of explanation; the question focuses on the inability to explain the decision, not whether the decision was biased. Option C is wrong because reliability and safety address system failures, robustness, and operational consistency, not the interpretability of model outputs. Option D is wrong because privacy and security deal with data protection, access controls, and confidentiality, not the explainability of model reasoning.

Full explanation →

286

MCQmedium

A self-driving car company develops an AI system that is highly accurate in testing but fails to consistently detect pedestrians during heavy rain. Which Microsoft responsible AI principle is most directly violated?

A.Fairness

B.Reliability and safety

C.Privacy and security

D.Transparency

AnswerB

This principle ensures AI systems operate safely and consistently; failing in adverse weather violates that.

Why this answer

The system fails to consistently detect pedestrians during heavy rain, which is a failure of the AI to perform reliably under real-world conditions. Microsoft's 'Reliability and safety' principle requires AI systems to operate dependably and safely across all expected scenarios, including edge cases like adverse weather. This directly violates that principle because the system's accuracy drops in a common environmental condition, posing safety risks.

Exam trap

Microsoft often tests the trap that candidates confuse 'reliability and safety' with 'fairness' when a system fails under specific conditions, but fairness is about demographic bias, not environmental robustness.

How to eliminate wrong answers

Option A is wrong because fairness concerns bias against protected groups (e.g., race, gender), not environmental conditions like rain. Option C is wrong because privacy and security involve data protection and unauthorized access, not operational performance in weather. Option D is wrong because transparency refers to explainability and disclosure of AI behavior, not the system's ability to function correctly under stress.

Full explanation →

287

MCQeasy

What is the role of a label (also called target or ground truth) in supervised machine learning?

A.A category of input features used by the model

B.The correct output or answer associated with each training example that the model learns to predict

C.A text description attached to a model explaining what it does

D.A tag applied to Azure ML resources for organization

AnswerB

Labels are the ground truth answers in training data — the model learns to produce predictions that match the labels.

Why this answer

In supervised machine learning, the label (also called target or ground truth) is the known correct output for each training example. The model uses these labels during training to learn the mapping from input features to outputs, enabling it to make accurate predictions on new, unseen data. This is fundamental to supervised learning, where the algorithm minimizes the error between its predictions and the ground truth labels.

Exam trap

The trap here is confusing the term 'label' in machine learning (ground truth output) with the general concept of a 'label' as a tag or category, leading candidates to mistakenly choose Option A or D.

How to eliminate wrong answers

Option A is wrong because a label is not a category of input features; input features are the independent variables used to make predictions, while the label is the dependent variable the model aims to predict. Option C is wrong because a text description attached to a model is documentation or metadata, not the ground truth used for training. Option D is wrong because a tag applied to Azure ML resources is an organizational metadata label for resource management, not a training data label used in supervised learning.

Full explanation →

288

MCQmedium

What is 'image segmentation' and how does it differ from object detection?

A.Dividing an image file into smaller files for distributed storage

B.Classifying every pixel in an image to identify precise boundaries — more detailed than bounding-box object detection

C.Removing the background from an image by detecting edges

D.Dividing the training dataset into segments for cross-validation

AnswerB

Segmentation goes beyond bounding boxes to pixel-level classification — enabling precise shape delineation rather than just location.

Why this answer

Image segmentation classifies every pixel in an image into a category, producing pixel-level masks that outline objects with precise boundaries. This differs from object detection, which only draws bounding boxes around objects and does not distinguish object edges or overlapping instances. Option B correctly captures this higher granularity and accuracy.

Exam trap

The trap here is that candidates confuse 'image segmentation' with simple background removal or edge detection, overlooking the requirement for pixel-level classification across all object categories.

How to eliminate wrong answers

Option A is wrong because it describes file splitting for storage, not a computer vision technique; image segmentation operates on pixel data within a single image, not on file distribution. Option C is wrong because it oversimplifies segmentation as mere background removal via edge detection, whereas true segmentation assigns every pixel to a class (e.g., road, car, pedestrian) and handles multiple objects and overlapping regions. Option D is wrong because it confuses dataset partitioning for model validation with the computer vision task of partitioning an image into semantic regions.

Full explanation →

289

MCQmedium

What is the purpose of Azure AI Content Safety in the context of generative AI deployments?

A.To compress generated content for faster delivery

B.To detect and filter harmful content in AI prompts and responses

C.To measure the quality and accuracy of AI-generated responses

D.To ensure AI content is written in the correct language

AnswerB

Content Safety screens generative AI inputs and outputs for violence, sexual content, hate speech, and other harmful categories.

Why this answer

Azure AI Content Safety is a service designed to detect and filter harmful content, such as hate speech, violence, self-harm, and sexually explicit material, in both user prompts and AI-generated responses. In generative AI deployments, this ensures that the model's outputs comply with safety policies and regulatory requirements, preventing the dissemination of offensive or dangerous content.

Exam trap

The trap here is that candidates confuse Azure AI Content Safety with general AI quality or language services, but the exam specifically tests its role as a safety filter for harmful content in generative AI pipelines, not for performance, accuracy, or language correctness.

How to eliminate wrong answers

Option A is wrong because Azure AI Content Safety does not perform compression; content delivery optimization is handled by services like Azure Content Delivery Network or Azure Front Door, not by a content safety filter. Option C is wrong because measuring quality and accuracy of AI responses is the role of evaluation metrics (e.g., BLEU, ROUGE, or Azure AI Studio's evaluation tools), not a safety detection service. Option D is wrong because language detection and translation are capabilities of Azure AI Translator or Azure AI Language, not Azure AI Content Safety, which focuses on harmful content regardless of language.

Full explanation →

290

MCQmedium

A marketing team uses Azure OpenAI Service to generate marketing copy. They notice the generated text is often repetitive, using the same phrases and words multiple times. Which parameter should they increase to directly reduce this repetition?

A.Temperature

B.Frequency penalty

C.Top-p

D.Max tokens

AnswerB

Correct. A higher frequency penalty reduces the likelihood of the model repeating the same tokens and phrases, directly addressing repetition.

Why this answer

Frequency penalty directly reduces repetition by penalizing tokens that have already appeared in the generated text. A higher frequency penalty value (e.g., 0.5 to 1.0) decreases the likelihood of the model reusing the same phrases or words, making the output more diverse and less repetitive.

Exam trap

The trap here is that candidates often confuse temperature or Top-p with repetition control, but those parameters affect randomness and diversity of vocabulary, not the direct penalization of repeated tokens that frequency penalty provides.

How to eliminate wrong answers

Option A is wrong because temperature controls the randomness of token selection, not repetition; increasing temperature makes output more random but does not specifically penalize repeated tokens. Option C is wrong because Top-p (nucleus sampling) limits the cumulative probability of token choices, affecting diversity of vocabulary but not directly penalizing repeated tokens. Option D is wrong because max tokens sets the maximum length of the generated response and has no effect on reducing repetition of phrases or words.

Full explanation →

291

MCQeasy

What is 'dimensionality reduction' and why is it useful in machine learning?

A.Reducing the physical size of AI hardware components for edge deployment

B.Reducing the number of input features while preserving key information for efficient modelling

C.Reducing the model's output to a single dimension for binary decision making

D.Simplifying the Azure ML workspace to have fewer compute resources and experiments

AnswerB

Dimensionality reduction (PCA, UMAP) removes redundant features — enabling faster training, less overfitting, and data visualisation.

Why this answer

Dimensionality reduction is the process of reducing the number of input features (variables) in a dataset while retaining as much of the original information as possible. This is useful in machine learning because it helps combat the 'curse of dimensionality', reduces overfitting, lowers computational cost, and can improve model performance by eliminating noise and redundant features. In Azure Machine Learning, techniques like Principal Component Analysis (PCA) are commonly used for this purpose.

Exam trap

The trap here is that candidates often confuse dimensionality reduction with model output simplification or hardware reduction, because the word 'reduction' is used broadly, but the exam specifically tests the definition as a feature preprocessing technique for input data.

How to eliminate wrong answers

Option A is wrong because it describes physical hardware downsizing for edge deployment, which is unrelated to the mathematical or algorithmic concept of reducing feature dimensions in a dataset. Option C is wrong because it confuses dimensionality reduction with collapsing the model's output to a single dimension for binary classification; dimensionality reduction applies to input features, not the output. Option D is wrong because it refers to simplifying an Azure ML workspace by reducing compute resources and experiments, which is an operational or administrative action, not a data preprocessing or feature engineering technique.

Full explanation →

292

MCQmedium

What is the difference between entities and intents in conversational language understanding?

A.Intents are for text; entities are for speech recognition

B.Intents represent the user's goal; entities are the specific pieces of information within the utterance

C.Intents are predefined answers; entities are user questions

D.They are the same concept with different names for clarity

AnswerB

Intent = what the user wants (BookFlight); Entity = the specific data in the utterance (Seattle, Tokyo, March 15).

Why this answer

In conversational language understanding (CLU), intents represent the user's overall goal or desired action (e.g., 'BookFlight'), while entities are specific data points extracted from the utterance that provide context for that intent (e.g., 'New York' as a destination). This distinction is fundamental to natural language processing (NLP) on Azure, where intents map to actions and entities provide the parameters needed to fulfill those actions.

Exam trap

The trap here is that candidates often confuse intents with responses or entities with questions, but the exam specifically tests the functional roles: intents classify the user's goal, while entities extract the specific data needed to act on that goal.

How to eliminate wrong answers

Option A is wrong because intents and entities are both used in text-based conversational language understanding, not limited to speech recognition; speech recognition is a separate Azure service (Speech-to-Text). Option C is wrong because intents are not predefined answers; they are classifications of user goals, while entities are not user questions but rather extracted pieces of information like dates or locations. Option D is wrong because intents and entities are distinct concepts with different roles in language understanding, not the same concept with different names.

Full explanation →

293

MCQmedium

A company uses Azure OpenAI Service to generate marketing copy. They notice that sometimes the generated text contains repetitive phrases or gets stuck in loops. They want to reduce this behavior without changing the overall creativity of the model. Which parameter should they adjust?

A.Increase the frequency_penalty parameter.

B.Decrease the temperature parameter.

C.Increase the presence_penalty parameter.

D.Decrease the top_p parameter.

AnswerA

Correct. Increasing frequency_penalty reduces the likelihood that the model will repeat the same tokens, making it effective against repetitive loops.

Why this answer

Increasing the frequency_penalty parameter reduces the likelihood of the model repeating the same phrases by penalizing tokens that have already appeared in the generated text. This directly addresses the repetitive loops without altering the overall creativity, as frequency_penalty specifically targets token frequency rather than randomness or diversity.

Exam trap

The trap here is that candidates often confuse frequency_penalty with presence_penalty, assuming both reduce repetition equally, but frequency_penalty specifically targets repeated occurrences while presence_penalty only discourages topic reuse.

How to eliminate wrong answers

Option B is wrong because decreasing temperature reduces randomness and makes the model more deterministic, which can actually increase repetition, not reduce it. Option C is wrong because presence_penalty penalizes tokens based on whether they have appeared at all, not their frequency, so it encourages new topics but does not specifically target repetitive phrases. Option D is wrong because decreasing top_p narrows the set of candidate tokens to the most probable ones, which reduces diversity and can worsen repetition, not fix it.

Full explanation →

294

MCQmedium

What is 'AI enrichment' in the context of Azure AI Search (Cognitive Search)?

A.Adding premium features to an Azure AI subscription

B.Applying AI cognitive skills during search indexing to extract and enrich content with metadata

C.Training custom ML models to improve search result ranking

D.Encrypting indexed search content with AI-managed keys

AnswerB

AI enrichment runs OCR, NER, sentiment, and other skills on indexed content — making unstructured data (images, PDFs) fully searchable.

Why this answer

AI enrichment in Azure AI Search refers to the process of applying built-in or custom cognitive skills during the indexing pipeline to extract, transform, and enrich unstructured data (e.g., images, text, PDFs) with additional metadata. This enables capabilities such as OCR, entity recognition, key phrase extraction, and language detection, turning raw content into searchable, structured information without requiring separate ML training.

Exam trap

The trap here is that candidates confuse AI enrichment (which extracts metadata during indexing) with custom ML model training for ranking or with general AI subscription features, leading them to select options that describe unrelated AI capabilities.

How to eliminate wrong answers

Option A is wrong because AI enrichment is not about adding premium features to an Azure AI subscription; it is a specific indexing capability within Azure AI Search that uses cognitive skills to enhance content. Option C is wrong because AI enrichment does not involve training custom ML models to improve search result ranking; ranking is handled by Azure AI Search's built-in scoring profiles and semantic search, not by enrichment skills. Option D is wrong because AI enrichment is unrelated to encryption; encryption of indexed content is managed via Azure Storage encryption or customer-managed keys, not through cognitive skills.

Full explanation →

295

MCQmedium

A game development studio uses Azure OpenAI Service to generate unique backstories for non-player characters (NPCs). They want the generated stories to be coherent and relevant to a given character class (e.g., warrior, mage) but also creative and varied. Which parameter should the studio adjust primarily to increase the creativity and variety of the generated text?

A.Increase the temperature parameter

B.Increase the top_p parameter

C.Increase the frequency_penalty parameter

D.Decrease the max_tokens parameter

AnswerA

Higher temperature values make the model's output more random and creative, leading to greater variety in generated backstories.

Why this answer

Increasing the temperature parameter makes the model's output more random by scaling the probability distribution over tokens, which encourages less likely word choices and thus increases creativity and variety in generated text. For the game studio, a higher temperature (e.g., 0.8–1.0) will produce more diverse and imaginative backstories for different character classes, while still maintaining coherence if not set too high.

Exam trap

Microsoft often tests the distinction between temperature and top_p, where candidates mistakenly think top_p is the primary creativity control, but temperature is the fundamental parameter for adjusting randomness and variety in text generation.

How to eliminate wrong answers

Option B is wrong because increasing top_p (nucleus sampling) also adds randomness but it does so by limiting the cumulative probability mass of token choices; while it can increase variety, it is not the primary parameter for controlling creativity—temperature is more direct. Option C is wrong because increasing frequency_penalty reduces the likelihood of repeating the same tokens or phrases, which can improve diversity but primarily targets repetition, not overall creativity or variety in story content. Option D is wrong because decreasing max_tokens limits the length of the output, which may reduce coherence and detail but does not inherently increase creativity or variety; it can even constrain the model's ability to generate varied content.

Full explanation →

296

MCQmedium

A company develops an AI-powered virtual assistant for customer service. To ensure the assistant can be used by people with visual impairments, the team integrates screen reader compatibility. Which Microsoft responsible AI principle is most directly addressed by this action?

A.Fairness

B.Reliability & Safety

C.Privacy & Security

D.Inclusiveness

AnswerD

Inclusiveness requires AI systems to be designed for all users, including those with disabilities.

Why this answer

Option D is correct because integrating screen reader compatibility directly addresses the inclusiveness principle of responsible AI. This principle ensures that AI systems are designed to be accessible and usable by people with diverse abilities, including those with visual impairments, by supporting assistive technologies like screen readers.

Exam trap

The trap here is that candidates may confuse inclusiveness with fairness, as both involve ethical considerations, but inclusiveness specifically targets accessibility for people with disabilities, while fairness addresses bias and discrimination across demographic groups.

How to eliminate wrong answers

Option A is wrong because fairness focuses on preventing bias and ensuring equitable treatment across different groups, not specifically on accessibility for people with disabilities. Option B is wrong because reliability and safety concern the system's ability to perform consistently and safely under various conditions, not its compatibility with assistive technologies. Option C is wrong because privacy and security involve protecting user data and ensuring secure interactions, which is unrelated to screen reader compatibility.

Full explanation →

297

Matchingmedium

Match each Azure AI concept to its definition.

Drag a concept onto its matching description — or click a concept then click the description.

Concepts

Matches

Probability that a prediction is correct

Coordinates around an object in an image

Identify main topics in text

Determine positive, negative, or neutral tone

Identify named entities like people or places

Why these pairings

These are fundamental concepts in AI-900.

Full explanation →

298

Multi-Selectmedium

A law firm needs to automatically process incoming legal documents. They have two specific requirements: (1) extract the names of all parties involved, the court name, and the filing date; (2) categorize each document as a 'complaint', 'motion', or 'subpoena'. Which two Azure AI Language features should they use? (Choose two.)

Select 2 answers

A.Sentiment analysis

B.Key phrase extraction

C.Custom text classification

D.Named entity recognition (NER)

AnswersC, D

Custom text classification can be trained to assign user-defined labels such as 'complaint', 'motion', or 'subpoena' to documents.

Why this answer

Custom text classification (C) is correct because it allows the law firm to train a model to categorize legal documents into custom classes like 'complaint', 'motion', or 'subpoena' based on labeled examples. This feature is designed for domain-specific classification tasks where predefined categories are insufficient.

Exam trap

The trap here is that candidates often confuse key phrase extraction with named entity recognition, but key phrase extraction returns untyped phrases rather than structured entities with predefined categories, and it cannot perform document-level classification.

Full explanation →

299

MCQhard

What is the bias-variance tradeoff in machine learning?

A.Choosing between model accuracy and computational cost

B.The balance between model simplicity (underfitting) and model complexity (overfitting)

C.Deciding whether to use biased training data or unbiased test data

D.The tradeoff between training speed and model size

AnswerB

High bias = underfitting (too simple); high variance = overfitting (too complex). The tradeoff finds optimal model complexity.

Why this answer

Option B is correct because the bias-variance tradeoff directly addresses the tension between underfitting (high bias, overly simple model) and overfitting (high variance, overly complex model). In Azure Machine Learning, this tradeoff is managed through hyperparameter tuning (e.g., regularization strength, tree depth) to achieve optimal generalization on unseen data.

Exam trap

The trap here is that candidates confuse 'bias' in the bias-variance tradeoff (model bias) with 'bias' in data fairness or ethical AI, leading them to incorrectly select Option C.

How to eliminate wrong answers

Option A is wrong because it confuses the bias-variance tradeoff with a resource allocation decision (accuracy vs. computational cost), which is a separate engineering concern, not a fundamental ML principle. Option C is wrong because it misrepresents bias as data bias (e.g., sampling bias) rather than model bias (systematic error from oversimplification), and the tradeoff involves variance, not test data selection. Option D is wrong because it conflates the tradeoff with operational metrics (training speed and model size), which are unrelated to the statistical concepts of bias and variance.

Full explanation →

300

MCQeasy

A photo sharing app wants to automatically generate descriptive captions for uploaded photos to improve accessibility for visually impaired users. Which Azure Computer Vision feature should they use?

A.Optical Character Recognition (OCR)

B.Object Detection

C.Image Classification

D.Describe Image (Image Captioning)

AnswerD

This feature generates a descriptive caption for an image, making it suitable for accessibility applications.

Why this answer

Option D is correct because the Describe Image (Image Captioning) feature of Azure Computer Vision generates human-readable captions that describe the content of an image. This directly meets the requirement of automatically generating descriptive captions for uploaded photos to improve accessibility for visually impaired users.

Exam trap

The trap here is that candidates often confuse Object Detection (identifying objects) with Image Captioning (describing the scene), or assume OCR is sufficient for accessibility when it only handles text extraction, not scene understanding.

How to eliminate wrong answers

Option A is wrong because Optical Character Recognition (OCR) extracts text from images, not descriptive captions about the image content. Option B is wrong because Object Detection identifies and locates specific objects within an image, but does not generate a natural language description of the overall scene. Option C is wrong because Image Classification assigns a single label or category to an image, not a multi-sentence descriptive caption.

Full explanation →

Microsoft Azure AI Fundamentals AI-900 (AI-900) — Questions 226–300