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

 

Question 321

A company wants to predict whether a customer will renew a subscription using previous customer records where renewal outcomes are known. Which learning approach should be used?

1. Unsupervised learning
2. Supervised learning
3. Reinforcement learning
4. Clustering

Correct Answer: 2

Explanation

Supervised learning uses historical examples that contain known target outcomes. In this scenario, previous customer records include information about customers along with known renewal results. The model can learn relationships between customer features and the renewal outcome, then use those learned patterns to predict whether new customers are likely to renew. Unsupervised learning does not use predefined target labels, clustering is a specific unsupervised technique, and reinforcement learning involves learning through interactions and rewards. Therefore, supervised learning is the appropriate approach for this prediction task.

Question 322

A retailer wants to group customers based on shopping behavior without knowing the groups in advance. Which machine learning technique can discover these groups?

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

Correct Answer: 3

Explanation

Clustering is an unsupervised learning technique that discovers groups within data based on similarities between observations. A retailer can provide customer information such as purchasing frequency, average spending, and product preferences without specifying the desired segments beforehand. The clustering algorithm can then identify naturally occurring groups. Classification requires predefined categories, regression predicts numerical values, and speech recognition processes spoken language. Therefore, clustering is appropriate when a retailer wants to discover customer segments without already knowing what those segments should be.

Question 323

A financial application needs to predict the exact amount of a customer’s monthly spending. Which machine learning task is most appropriate?

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

Correct Answer: 3

Explanation

Regression is used when a machine learning model needs to predict a numerical value. Monthly customer spending is a quantitative target, so a regression model can learn relationships between relevant features and historical spending amounts. The model could use information such as previous purchases, account activity, or customer characteristics to estimate a future spending value. Classification predicts categories, clustering identifies groups without predefined labels, and entity recognition extracts entities from text. Because the required output is a numerical spending amount, regression is the appropriate machine learning task.

Question 324

An application receives a photograph of a document and needs to convert the visible printed words into machine-readable text. Which capability should it use?

1. OCR
2. Text-to-speech
3. Clustering
4. Sentiment analysis

Correct Answer: 1

Explanation

Optical character recognition, or OCR, extracts written characters from images and converts them into machine-readable text. It can be used for photographs of receipts, forms, signs, scanned documents, and other visual content containing text. Text-to-speech converts written text into spoken audio, clustering groups similar data, and sentiment analysis evaluates opinions expressed in language. Since the application starts with an image containing printed words and needs editable or machine-readable text as the output, OCR is the appropriate capability.

Question 325

A virtual assistant needs to understand questions such as “Where is my order?” and “Can you tell me my delivery status?” Which AI workload is most relevant?

1. Computer vision
2. Natural language processing
3. Regression
4. Image classification

Correct Answer: 2

Explanation

Natural language processing enables applications to analyze and interpret human language. A virtual assistant can use NLP capabilities to understand different ways users express similar requests and determine the relevant intent. In this example, both questions concern order delivery status even though the wording differs. Computer vision and image classification process visual information, while regression predicts numerical values. Therefore, natural language processing is the appropriate AI workload for understanding and responding to written or spoken user questions.

Question 326

A company wants to automatically identify the emotional tone of messages submitted to its customer service department. Which capability is appropriate?

1. OCR
2. Object detection
3. Sentiment analysis
4. Speech synthesis

Correct Answer: 3

Explanation

Sentiment analysis evaluates written language to determine the expressed sentiment or emotional tone. Customer service messages can be analyzed to identify whether customers are expressing positive, negative, or neutral sentiment. This information can help organizations identify recurring issues, monitor customer satisfaction, and prioritize responses. OCR extracts text from images, object detection identifies visual objects, and speech synthesis converts written content into spoken audio. Because the input consists of customer messages and the requirement is to determine their sentiment, sentiment analysis is the appropriate language capability.

Question 327

A company wants to create a system that identifies whether an image contains a cat, dog, car, or bicycle, with one or more categories assigned to the image. Which computer vision task is appropriate?

1. Image classification
2. Speech-to-text
3. Key phrase extraction
4. Regression

Correct Answer: 1

Explanation

Image classification assigns one or more categories or labels to an image based on its visual content. It is appropriate when the primary requirement is to determine what category or categories an image belongs to rather than locating individual objects within it. Speech-to-text converts spoken audio into written text, key phrase extraction identifies important concepts in text, and regression predicts numerical values. If the application needs to classify images according to categories such as cat, dog, car, or bicycle, image classification is the appropriate computer vision task.

Question 328

A manufacturer wants to identify every damaged component in an image and determine the location of each damaged component. Which capability is most suitable?

1. Image classification
2. Object detection
3. Sentiment analysis
4. Language detection

Correct Answer: 2

Explanation

Object detection identifies objects within images and provides information about their locations. A manufacturing inspection system can use it to locate damaged components and potentially identify their categories or characteristics. Image classification can determine the category of an entire image but does not primarily provide precise locations for multiple objects. Sentiment analysis and language detection operate on text rather than visual components. Therefore, object detection is more suitable when the application must both identify damaged components and determine where they appear within an image.

Question 329

An organization wants to convert a written training script into spoken audio for an instructional application. Which capability should be used?

1. Speech-to-text
2. Text-to-speech
3. OCR
4. Entity recognition

Correct Answer: 2

Explanation

Text-to-speech converts written text into spoken audio. A training application can provide a written script to a speech synthesis service and generate an audio version that users can listen to. Speech-to-text performs the opposite conversion by turning spoken language into written text. OCR extracts text from images, while entity recognition identifies meaningful entities in written content. Because the source material is a written training script and the desired output is spoken audio, text-to-speech is the appropriate AI capability.

Question 330

A company wants to translate spoken customer questions into another language while preserving the information in the spoken message. Which combination can support the workflow?

1. Speech-to-text and translation
2. OCR and object detection
3. Regression and clustering
4. Image classification and sentiment analysis

Correct Answer: 1

Explanation

Speech-to-text can first convert the spoken customer question into written text. A translation capability can then translate that text into the required target language. This workflow allows an application to process spoken questions even when the user and recipient communicate in different languages. OCR and object detection are visual capabilities, while regression and clustering are machine learning techniques for different purposes. Image classification and sentiment analysis also do not provide the required speech-to-language translation workflow. Therefore, speech-to-text followed by translation is appropriate.

Question 331

A document-processing system must extract standard information from commonly supported document types without training a new model. Which option can simplify the implementation?

1. A prebuilt model
2. A clustering algorithm
3. A regression model
4. Speech recognition

Correct Answer: 1

Explanation

A prebuilt model can simplify document-processing workloads when the document type and required fields are already supported by an available service. Instead of collecting and labeling training examples to create a custom model, an organization can use an existing model designed for the relevant document scenario. Clustering groups similar data, regression predicts numerical values, and speech recognition processes spoken audio. Therefore, when standard document information can be handled by an existing service model, a prebuilt model can reduce the development effort required.

Question 332

A company wants to extract specific fields from a specialized form that has a unique layout used only by the organization. Which approach is more suitable than relying entirely on a standard prebuilt model?

1. Custom document model
2. Text-to-speech
3. Sentiment analysis
4. Language detection

Correct Answer: 1

Explanation

A custom document model can be trained for specialized organizational forms when standard prebuilt models do not adequately support the document structure or required fields. The organization can provide representative examples and identify the information it wants to extract. This allows the model to be tailored to the organization’s specific document format and business requirements. Text-to-speech generates spoken audio, sentiment analysis evaluates opinions, and language detection identifies written languages. Therefore, a custom document model is appropriate for unique forms requiring specialized field extraction.

Question 333

A machine learning team wants to use an automated process to identify suitable algorithms and model configurations for a dataset instead of manually testing every possible combination. Which Azure capability can help?

1. Azure Machine Learning automated ML
2. Azure AI Speech
3. Azure AI Vision
4. Azure AI Translator

Correct Answer: 1

Explanation

Automated ML in Azure Machine Learning can automate parts of the model-development process by testing different algorithms, configurations, and approaches against a dataset to help identify suitable models. It can assist with tasks such as model selection and performance comparison while reducing the amount of manual experimentation required. Azure AI Speech focuses on speech workloads, Azure AI Vision handles visual information, and Azure AI Translator performs translation. Therefore, Azure Machine Learning automated ML is the relevant capability for automating model experimentation and selection.

Question 334

A team wants to organize multiple machine learning steps, such as data preparation, training, and evaluation, into a repeatable workflow. Which capability is most appropriate?

1. Machine learning pipeline
2. OCR
3. Sentiment analysis
4. Image captioning

Correct Answer: 1

Explanation

A machine learning pipeline organizes multiple processing steps into a structured and repeatable workflow. Steps can include data preparation, feature processing, model training, evaluation, and other operations required by the solution. Using a pipeline can make workflows easier to manage, reproduce, and automate. OCR extracts text from images, sentiment analysis evaluates text sentiment, and image captioning generates descriptions of images. Therefore, when a team needs to connect several machine learning stages into an organized process, a machine learning pipeline is the appropriate capability.

Question 335

A generative AI application produces a response that contains information not supported by the available source material. Which issue should the development team investigate?

1. Hallucination
2. Clustering
3. Feature scaling
4. Object detection

Correct Answer: 1

Explanation

A hallucination occurs when a generative AI model produces information that is inaccurate, fabricated, or unsupported by the available evidence. This can be especially problematic when an application is expected to answer using trusted organizational information. Developers can use grounding, retrieval-augmented generation, careful prompting, validation, and evaluation to help reduce unsupported responses. Clustering groups data, feature scaling adjusts numerical feature ranges, and object detection identifies visual objects. Therefore, an unsupported generated statement should be investigated as a potential hallucination.

Question 336

An AI application retrieves information from a company’s document repository before asking a language model to generate a response. Which architecture does this describe?

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

Correct Answer: 1

Explanation

Retrieval-augmented generation combines information retrieval with generative AI. The application first searches an external source, such as a company document repository, and retrieves relevant information. That information is then supplied to the language model as context for generating an answer. This architecture can help applications respond using current or organization-specific information without depending solely on the model’s original training knowledge. Image classification, regression, and speech synthesis address different workloads. Therefore, retrieving company documents before generating a response describes retrieval-augmented generation.

Question 337

A development team wants to monitor whether a deployed model’s predictions remain within expected performance levels over time. What activity should the team perform?

1. Model monitoring
2. Image captioning
3. OCR
4. Translation

Correct Answer: 1

Explanation

Model monitoring involves observing a deployed AI or machine learning system to determine whether its behavior and performance remain within expected limits. Monitoring can help identify changes in input data, prediction patterns, performance metrics, or other indicators that may require investigation. A model that performs well during initial evaluation can behave differently after deployment as real-world conditions change. Image captioning, OCR, and translation are AI capabilities rather than deployment-monitoring activities. Therefore, ongoing model monitoring is appropriate for tracking whether a deployed model continues to meet expectations.

Question 338

An AI system handles personal customer information and the organization wants to limit access to that information to authorized users. Which practice is most relevant?

1. Access control
2. Image classification
3. Temperature adjustment
4. Clustering

Correct Answer: 1

Explanation

Access control helps ensure that only authorized users or systems can access protected information. For AI applications that process personal customer data, appropriate access controls can reduce the risk of unauthorized disclosure or misuse. Organizations may combine access controls with other security measures such as encryption, identity management, and appropriate data-handling policies. Image classification analyzes visual data, temperature influences generative model variability, and clustering groups similar observations. Because the requirement concerns restricting access to personal information, access control is directly relevant.

Question 339

A classification model has both precision and recall requirements, and the team wants one metric that combines these two measures. Which metric should be considered?

1. Accuracy
2. Mean absolute error
3. F1 score
4. Regression error

Correct Answer: 3

Explanation

The F1 score combines precision and recall into a single metric by calculating their harmonic mean. It can be useful when a classification problem requires consideration of both the correctness of positive predictions and the ability to identify actual positive cases. Accuracy provides the overall proportion of correct classifications, while mean absolute error is generally used for numerical prediction. Regression error is also associated with numerical prediction rather than classification. Therefore, when a team specifically wants a combined measure of precision and recall, the F1 score is appropriate.

Question 340

Before deploying an AI model, a team compares its measured results with predefined business requirements and relevant evaluation metrics. What activity are they performing?

1. Model evaluation
2. Speech synthesis
3. Image segmentation
4. Data translation

Correct Answer: 1

Explanation

Model evaluation involves measuring a model’s performance against appropriate criteria and determining whether it meets the requirements of the intended application. Depending on the task, the team may examine metrics such as accuracy, precision, recall, F1 score, or mean absolute error. Comparing these results with predefined business and technical requirements helps determine whether additional improvement or testing is necessary before deployment. Speech synthesis, image segmentation, and translation are AI capabilities for different workloads. Therefore, comparing measured model performance with predefined requirements is model evaluation.