View Full Microsoft AI-900 Exam Dumps and Practice Test Dumps.

 

Question 281

A company wants to analyze images from security cameras to determine whether specific types of vehicles are present. Which AI workload is most appropriate?

1. Natural language processing
2. Speech processing
3. Computer vision
4. Regression

Correct Answer: 3

Explanation

Computer vision is designed to process and interpret visual information from images and video. A security-camera application can use computer vision capabilities to identify vehicles and determine relevant characteristics in captured frames. Depending on the exact requirement, image classification or object detection could be selected. Natural language processing handles written or spoken language, speech processing focuses on audio, and regression predicts numerical values. Since the input consists of camera images and the application needs to identify visual objects, computer vision is the appropriate AI workload.

Question 282

A machine learning model is trained using examples that contain both customer information and a known outcome showing whether each customer canceled a subscription. What does the known outcome represent?

1. Feature
2. Label
3. Prediction
4. Cluster

Correct Answer: 2

Explanation

A label represents the known target value associated with a training example in supervised learning. In this scenario, whether a customer canceled a subscription is the outcome the model is expected to learn to predict. Customer information such as age, usage, or purchase history can serve as features used to make that prediction. A prediction is the output generated by the trained model, while a cluster represents a group discovered through unsupervised learning. Therefore, the known cancellation outcome is the label.

Question 283

A business wants to estimate the amount of electricity a building will consume tomorrow based on historical consumption and weather information. Which machine learning task is appropriate?

1. Classification
2. Clustering
3. Regression
4. Entity recognition

Correct Answer: 3

Explanation

Regression is appropriate when the desired prediction is a numerical value. Electricity consumption can be represented as a measurable quantity, such as kilowatt-hours, and historical usage and weather conditions can be used as features. The model can learn relationships between these inputs and future consumption values. Classification predicts predefined categories, clustering discovers groups without known labels, and entity recognition extracts meaningful entities from text. Although the scenario involves a future value, the key requirement is predicting a numerical quantity, making regression the appropriate machine learning task.

Question 284

An organization has thousands of product descriptions and wants to automatically identify important terms that summarize the main topics in each description. Which capability should it use?

1. Key phrase extraction
2. Object detection
3. Speech synthesis
4. Image classification

Correct Answer: 1

Explanation

Key phrase extraction identifies important words or phrases that represent the main concepts in a piece of text. A company can use it to process product descriptions and obtain meaningful terms that summarize their content. This can support search, categorization, analytics, and content organization. Object detection analyzes visual objects, speech synthesis converts written text into spoken audio, and image classification assigns categories to images. Because the requirement involves identifying important concepts within written product descriptions, key phrase extraction is the appropriate natural language capability.

Question 285

A developer needs to convert a recorded customer conversation into written text so that it can be searched and analyzed later. Which capability should be used?

1. Text-to-speech
2. Speech-to-text
3. Image captioning
4. OCR

Correct Answer: 2

Explanation

Speech-to-text converts spoken audio into written text. A recorded customer conversation can therefore be transcribed and stored as text, allowing the organization to search the conversation and perform additional analysis such as sentiment detection or key phrase extraction. Text-to-speech performs the reverse process by converting written content into spoken audio. Image captioning describes visual content, while OCR extracts text from images or scanned documents. Since the input is a recorded conversation and the required output is written text, speech-to-text is the appropriate capability.

Question 286

A retailer wants to determine which products appear in a photograph and where each product is located within the image. Which capability is most suitable?

1. Sentiment analysis
2. Language detection
3. Object detection
4. Text summarization

Correct Answer: 3

Explanation

Object detection identifies objects within an image and determines their locations, often by providing bounding boxes around detected objects. This makes it suitable for a retail application that needs to identify products and determine where they appear in a photograph. Sentiment analysis evaluates opinions expressed in text, language detection identifies the language of written content, and text summarization creates shorter versions of longer text. Because the requirement involves both recognizing visual objects and locating them within an image, object detection is the appropriate capability.

Question 287

A generative AI application needs to answer questions using information stored in a company’s internal documents. The documents may change regularly. Which approach can help the application use the latest available information?

1. Retrieval-augmented generation
2. Image classification
3. Regression
4. Speech recognition

Correct Answer: 1

Explanation

Retrieval-augmented generation allows an application to retrieve relevant information from an external knowledge source before generating a response. This is useful when organizational documents change regularly because the application can retrieve current content rather than depending exclusively on information learned during the model’s original training. The retrieved information can then be provided as context to the generative model. Image classification, regression, and speech recognition address different AI workloads. Therefore, retrieval-augmented generation is suitable for answering questions using frequently updated organizational information.

Question 288

A company wants to reduce the number of legitimate transactions incorrectly flagged as fraudulent. Which classification outcome should it focus on reducing?

1. True positives
2. False negatives
3. False positives
4. True negatives

Correct Answer: 3

Explanation

A false positive occurs when a model predicts that a condition exists when it actually does not. In fraud detection, a false positive means a legitimate transaction is incorrectly identified as fraudulent. Reducing false positives can therefore help prevent unnecessary transaction reviews, customer interruptions, or declined legitimate purchases. True positives are correctly identified fraudulent transactions, false negatives are fraudulent transactions that were missed, and true negatives are legitimate transactions correctly recognized as legitimate. Since the organization wants fewer legitimate transactions incorrectly flagged, false positives are the relevant outcome.

Question 289

A generative AI system should produce more consistent responses for a business application where predictability is important. Which temperature setting generally supports this goal?

1. A higher temperature
2. A lower temperature
3. A random temperature for every request
4. No temperature setting

Correct Answer: 2

Explanation

A lower temperature generally encourages more predictable and consistent generative AI outputs by reducing variation in token selection. This can be useful for applications where responses need to follow a stable pattern or where creativity is less important than consistency. A higher temperature generally increases variation and can be useful for brainstorming or creative tasks. Randomly changing the temperature makes behavior less predictable, while removing the setting does not specifically provide greater consistency. Therefore, a lower temperature is generally more suitable when predictable responses are desired.

Question 290

A company receives standardized identification documents and wants to use an Azure service that already understands common fields in those document types. Which option is most appropriate?

1. A prebuilt Document Intelligence model
2. A clustering algorithm
3. Speech recognition
4. Image captioning

Correct Answer: 1

Explanation

Prebuilt models in Azure AI Document Intelligence are designed to process supported common document types and extract relevant information without requiring an organization to create a model entirely from scratch. This can simplify document-processing scenarios involving standardized documents with commonly recognized fields. A custom model may be appropriate when the organization’s documents require specialized extraction behavior. Clustering groups data, speech recognition processes spoken audio, and image captioning generates descriptions of images. Therefore, a prebuilt Document Intelligence model is appropriate when the document type is already supported.

Question 291

An organization wants users to understand the limitations and intended use of an AI system before relying on its outputs. Which responsible AI principle is most relevant?

1. Transparency
2. Clustering
3. Regression
4. Speech synthesis

Correct Answer: 1

Explanation

Transparency involves providing appropriate information about an AI system so users and stakeholders can understand its capabilities, limitations, intended use, and relevant behavior. Clear documentation and communication can help users interpret AI-generated results appropriately and avoid relying on a system outside its intended purpose. Clustering and regression are machine learning techniques, while speech synthesis converts text into spoken audio. Since the scenario focuses on helping users understand how and where an AI system should be used, transparency is the most relevant responsible AI principle.

Question 292

An AI application should support users who have different physical abilities and may interact with technology in different ways. Which responsible AI principle should guide the design?

1. Fairness
2. Inclusiveness
3. Privacy
4. Reliability

Correct Answer: 2

Explanation

Inclusiveness focuses on ensuring that AI systems can be used effectively by people with different abilities, backgrounds, and circumstances. An inclusive design may consider accessibility requirements and different ways users interact with technology. Fairness is concerned with equitable treatment and outcomes, privacy focuses on protecting personal information, and reliability addresses dependable system behavior. Because the scenario specifically emphasizes supporting users with different physical abilities and interaction needs, inclusiveness is the responsible AI principle that most directly addresses the requirement.

Question 293

A machine learning model uses features measured in very different numerical ranges. Why might preprocessing be needed before training?

1. To translate the features into another language
2. To make relevant numerical features more comparable in scale
3. To convert all data into images
4. To remove every categorical value automatically

Correct Answer: 2

Explanation

Numerical features can have very different ranges, such as age measured in tens and income measured in thousands. For algorithms that are sensitive to feature magnitude, scaling can help place relevant numerical features on more comparable ranges. This can improve how some models learn from the available data. Feature scaling does not translate information, convert all data into images, or automatically remove every categorical value. Therefore, preprocessing may be used to make numerical features more comparable and prevent differences in scale from disproportionately affecting model behavior.

Question 294

A deployed AI model’s performance gradually decreases because the real-world input data has changed significantly from the data used during development. What should the team investigate?

1. Data drift
2. OCR
3. Text synthesis
4. Image resolution

Correct Answer: 1

Explanation

Data drift occurs when the characteristics or distribution of input data change over time compared with the data used during model development. If real-world inputs become substantially different, a model may no longer perform as effectively as it did during evaluation. Monitoring input data and model performance can help identify these changes and determine whether retraining or other adjustments are needed. OCR extracts text from images, text synthesis is related to generating language, and image resolution concerns visual quality. Therefore, data drift should be investigated when deployed input patterns change.

Question 295

A regression model predicts house prices. The average absolute difference between its predictions and actual prices is required. Which metric should be used?

1. Accuracy
2. Recall
3. Mean absolute error
4. Precision

Correct Answer: 3

Explanation

Mean absolute error measures the average absolute difference between predicted and actual numerical values. It is commonly used to evaluate regression models, including models that predict house prices. A lower MAE indicates that predictions are, on average, closer to the actual values. Accuracy, precision, and recall are primarily associated with classification evaluation rather than measuring the average numerical prediction error in a regression problem. Therefore, mean absolute error is the appropriate metric when a team wants to determine the average absolute difference between predicted and actual house prices.

Question 296

A company wants to determine whether incoming emails are legitimate or spam. The model must select one of these two categories. Which task is being performed?

1. Regression
2. Classification
3. Clustering
4. Forecasting

Correct Answer: 2

Explanation

Classification assigns observations to predefined categories. In this scenario, every incoming email must be assigned to either the legitimate or spam category, making it a binary classification problem. The model can be trained using labeled examples of previous emails and then use learned patterns to classify new messages. Regression predicts numerical values, clustering discovers groups without predefined categories, and forecasting predicts future values over time. Because the output consists of two predefined categories, classification is the appropriate machine learning task.

Question 297

A generative AI developer wants responses to follow a specific format, such as returning three bullet points followed by a short recommendation. What can help communicate this requirement?

1. Clear prompt instructions
2. Image segmentation
3. Batch inference
4. Feature scaling

Correct Answer: 1

Explanation

Clear prompt instructions can tell a generative AI model how the response should be structured. Developers can specify requirements such as the number of items, formatting style, length, tone, or information that must be included. Clear instructions can make the expected output easier for the model to follow and easier for users to consume. Image segmentation processes visual regions, batch inference handles groups of predictions, and feature scaling preprocesses numerical data. Therefore, clear prompt instructions are the appropriate approach for requesting a specific response format.

Question 298

An organization wants to protect personal information contained in customer records used by an AI system. Which responsible AI principle is directly relevant?

1. Inclusiveness
2. Transparency
3. Privacy and security
4. Clustering

Correct Answer: 3

Explanation

Privacy and security focus on protecting information and ensuring that data is handled appropriately throughout an AI system’s lifecycle. Customer records may contain personal or sensitive information, so organizations should consider appropriate access controls, data protection measures, and responsible data-handling practices. Inclusiveness addresses broad accessibility and participation, while transparency concerns communicating appropriate information about AI systems. Clustering is a machine learning technique and is not a responsible AI principle. Because the requirement specifically concerns protecting personal information, privacy and security are directly relevant.

Question 299

A team creates a model that performs well on its training examples but poorly on new examples. They want to determine whether the model has learned the training data too specifically. What problem should they investigate?

1. Underfitting
2. Overfitting
3. Translation
4. Entity extraction

Correct Answer: 2

Explanation

Overfitting occurs when a machine learning model learns the training examples too specifically and does not generalize well to unseen data. A common indicator is high performance on the training dataset combined with substantially lower performance on evaluation or test data. Teams can investigate the model, training process, features, data volume, and regularization strategies when addressing potential overfitting. Underfitting usually results in poor performance even on training data. Translation and entity extraction are language capabilities and do not describe this machine learning problem.

Question 300

An AI project team wants to determine whether its model meets predefined performance requirements before deployment. Which activity is most appropriate?

1. Model evaluation
2. Image compression
3. Speech synthesis
4. Data encryption

Correct Answer: 1

Explanation

Model evaluation involves measuring a trained model against appropriate performance criteria using suitable evaluation data and metrics. The team can determine whether the model meets requirements for accuracy, precision, recall, error, or other relevant measures depending on the workload. Evaluation before deployment helps identify models that may require improvement or additional testing. Image compression reduces file size, speech synthesis generates spoken audio, and data encryption protects information. Since the project team needs to determine whether predefined model performance requirements have been met, model evaluation is the appropriate activity.