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

 

Question 101

A company wants to automatically extract fields such as invoice number, vendor name, and total amount from many invoices. Which Azure service is most appropriate for this requirement?

1. Azure AI Vision
2. Azure AI Speech
3. Azure AI Language
4. Azure AI Document Intelligence

Correct Answer: 4

Explanation

Azure AI Document Intelligence is designed to analyze documents and extract structured information from their content. It can process invoices, receipts, forms, and other business documents using prebuilt or custom models. For invoices, a prebuilt invoice model can identify fields such as vendor information, invoice numbers, dates, and totals. Azure AI Vision focuses primarily on image analysis, while Speech handles audio and Language provides text-based language capabilities. Therefore, Document Intelligence is the suitable service when an application needs structured data extracted from business documents.

Question 102

Which type of machine learning data contains values such as age, salary, and number of purchases?

1. Numerical data
2. Image data
3. Audio data
4. Text data

Correct Answer: 1

Explanation

Numerical data consists of values represented by numbers and can be used directly or transformed for machine learning purposes. Examples include a customer’s age, annual salary, number of purchases, temperature, and product price. These values can often be analyzed mathematically and used as features when training a model. Image data contains visual information, audio data represents sound, and text data contains written language. Recognizing the data type helps determine suitable preprocessing methods and machine learning techniques for a particular AI solution.

Question 103

An organization wants to build an application that searches a large collection of documents and returns relevant information based on a user’s query. Which Azure capability can support this type of solution?

1. Azure AI Speech
2. Azure AI Vision
3. Azure AI Search
4. Azure AI Face

Correct Answer: 3

Explanation

Azure AI Search provides capabilities for indexing and searching large collections of information. It can support applications that need to retrieve relevant documents or information based on user queries. AI enrichment capabilities can also help process content and make additional information available for searching. This makes it useful for knowledge mining and enterprise search scenarios. Azure AI Speech is focused on spoken language, while Vision handles visual analysis. Therefore, Azure AI Search is appropriate when the primary requirement is discovering relevant information within a large searchable content collection.

Question 104

A chatbot repeatedly produces overly creative answers when users ask factual questions. Which generative AI setting can be adjusted to make the responses more predictable?

1. Token limit
2. Temperature
3. Image resolution
4. Training dataset size

Correct Answer: 2

Explanation

Temperature controls the degree of randomness used when a generative AI model selects its next output tokens. A higher temperature can produce more varied and creative responses, while a lower temperature generally makes responses more predictable and consistent. For a chatbot that needs to provide factual answers with less variation, lowering the temperature can be useful. Token limits control how much content can be generated rather than directly controlling randomness. Image resolution and training dataset size do not serve this purpose during response generation.

Question 105

Which Azure AI Document Intelligence approach should be considered when a company has a unique document format that is not adequately handled by a prebuilt model?

1. A custom model
2. A speech model
3. A clustering model
4. A translation model

Correct Answer: 1

Explanation

A custom Document Intelligence model is useful when an organization needs to extract information from documents with formats or fields that are specific to its business. Custom models can be trained using representative sample documents so the service can learn where relevant information appears. Prebuilt models are convenient for common document types such as invoices and receipts, but they may not cover every specialized format. A speech, clustering, or translation model addresses a different type of AI workload and would not directly solve custom document extraction requirements.

Question 106

A developer creates a generative AI application that allows users to ask questions about an organization’s internal policies. The application retrieves relevant policy documents before sending context to the language model. What technique is being used?

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

Correct Answer: 4

Explanation

Retrieval-augmented generation, or RAG, combines information retrieval with generative AI. Before generating an answer, the application retrieves relevant information from an external knowledge source and provides that information as context to the language model. This approach can help the model produce answers that are grounded in an organization’s current documents rather than relying only on information learned during model training. It is particularly useful for internal knowledge assistants, document-based question answering, and enterprise chat applications where responses should reference a specific collection of information.

Question 107

Which metric focuses on the proportion of positive predictions that are actually positive?

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

Correct Answer: 2

Explanation

Precision measures how many of the instances predicted as positive are actually positive. It is calculated by dividing true positives by the total number of predicted positives, which includes both true positives and false positives. Precision is particularly important when false positive results are costly. Recall measures the proportion of actual positive cases that were successfully identified, while accuracy considers correct predictions across all classes. Mean absolute error is generally used for regression problems. Understanding these metrics helps evaluate classification models according to the business requirement.

Question 108

A bank wants an AI system to identify suspicious transactions that differ significantly from a customer’s normal behavior. Which AI workload is most relevant?

1. Image analysis
2. Natural language processing
3. Anomaly detection
4. Speech synthesis

Correct Answer: 3

Explanation

Anomaly detection is used to identify unusual patterns or observations that differ from expected behavior. In banking, a model can analyze transaction characteristics such as amount, location, timing, and frequency to identify activity that appears unusual compared with established patterns. The result can then be reviewed or used as part of a fraud detection process. Image analysis focuses on visual data, natural language processing handles human language, and speech synthesis generates spoken audio. Therefore, anomaly detection is the appropriate workload for identifying unusual transaction behavior.

Question 109

A generative AI application has a fixed maximum number of tokens that it can process in a single request. What does this limitation primarily affect?

1. The color of generated images
2. The model’s network bandwidth
3. The user’s authentication method
4. The amount of text that can be included or generated

Correct Answer: 4

Explanation

Tokens are units used by language models to represent portions of text. A model’s token limit determines how much input and output content can be handled within a request or conversation context. If a request contains a large document, the available token capacity may limit how much information can be supplied to the model or how much content can be generated. Token limits do not determine image colors, authentication methods, or network bandwidth. Understanding token limitations is important when designing applications that process long documents or conversations.

Question 110

A company wants to classify customer support messages into categories such as billing, technical support, shipping, and account access. Which Azure AI Language capability is relevant?

1. Custom text classification
2. Optical character recognition
3. Speech synthesis
4. Object detection

Correct Answer: 1

Explanation

Custom text classification in Azure AI Language can categorize text according to categories defined for a specific business requirement. A company could train a custom classification project using examples of support messages labeled as billing, technical support, shipping, or account access. Once trained, the model can classify new messages into the appropriate categories. OCR is used to extract text from images, speech synthesis converts text into spoken audio, and object detection identifies objects in images. Therefore, custom text classification directly matches the customer-support categorization scenario.

Question 111

Which type of data would be most appropriate for predicting electricity demand at different times throughout the day and across multiple dates?

1. Image data
2. Time-series data
3. Audio data
4. Categorical labels only

Correct Answer: 2

Explanation

Time-series data consists of observations collected over time, often at regular intervals. Electricity demand measured every hour, for example, forms a time series because each observation is associated with a particular point in time. Historical demand can be analyzed to identify trends, recurring patterns, and seasonal behavior that may help predict future consumption. Image and audio data represent different modalities, while categorical labels alone do not capture the temporal sequence required for this type of forecasting problem. Time-series information is therefore appropriate for electricity-demand prediction.

Question 112

An application receives a user’s spoken question and needs to return the same question translated into another language as spoken audio. Which combination of capabilities is required?

1. Image analysis and OCR
2. Sentiment analysis and classification
3. Speech recognition, translation, and speech synthesis
4. Object detection and image classification

Correct Answer: 3

Explanation

This scenario requires several language and speech capabilities working together. Speech recognition converts the user’s spoken question into text. Translation converts that recognized content into the target language. Speech synthesis then converts the translated text back into spoken audio. Combining these capabilities allows an application to accept spoken input and provide spoken output in another language. Image analysis, OCR, object detection, and image classification are designed for visual workloads and therefore do not provide the complete set of capabilities required for this speech translation scenario.

Question 113

A machine learning engineer removes duplicate records and handles missing values before training a model. What general stage of machine learning is being performed?

1. Data preprocessing
2. Model deployment
3. Inference
4. Model monitoring

Correct Answer: 1

Explanation

Data preprocessing involves preparing raw data so it can be used effectively for machine learning. Common activities include removing duplicate records, handling missing values, correcting inconsistent entries, transforming values, and preparing features in suitable formats. High-quality preprocessing can improve the reliability of the resulting model because poor or inconsistent input data can negatively affect learning. Model deployment occurs when a trained model is made available for applications, inference involves generating predictions, and monitoring evaluates a deployed solution. Therefore, cleaning and preparing raw records is part of data preprocessing.

Question 114

A generative AI application must prevent users from receiving responses containing prohibited or harmful content. Which capability should be included as part of the solution?

1. Regression analysis
2. Data normalization
3. Clustering
4. Content safety controls

Correct Answer: 4

Explanation

Content safety controls help identify and manage potentially harmful or inappropriate content in AI applications. They can be used to detect categories of problematic input or generated output and support policies for filtering, blocking, or reviewing responses. This is especially important when generative AI applications are exposed directly to users because prompts and generated responses can contain undesirable material. Regression, normalization, and clustering address machine learning data or modeling tasks rather than content moderation. A responsible generative AI solution should therefore include suitable content safety mechanisms.

Question 115

Which situation is an example of a classification problem rather than a regression problem?

1. Predicting tomorrow’s temperature
2. Predicting whether an email is spam or not spam
3. Estimating the selling price of a house
4. Forecasting monthly electricity consumption

Correct Answer: 2

Explanation

Classification predicts a category or class rather than a continuous numerical value. Determining whether an email is spam or not spam produces one of two categories, making it a binary classification problem. In contrast, predicting temperature, house prices, or electricity consumption generally produces numerical values and is therefore associated with regression or forecasting approaches. Correctly identifying the type of prediction required is important when selecting a machine learning algorithm, preparing evaluation metrics, and interpreting the resulting model.

Question 116

A vision application assigns descriptive tags such as “car,” “tree,” and “building” to an image without needing to locate the exact position of each object. Which capability best matches this requirement?

1. Object detection
2. Face verification
3. Speech recognition
4. Image tagging

Correct Answer: 4

Explanation

Image tagging identifies concepts or objects present in an image and associates descriptive tags with the image. For example, an image containing a car, tree, and building might receive tags corresponding to those concepts. Object detection goes further by identifying objects and providing their locations within the image, usually with bounding regions. Face verification focuses on comparing facial identities, while speech recognition processes spoken language. Therefore, image tagging is the appropriate capability when the requirement is to describe the contents of an image without needing precise object locations.

Question 117

A model performs extremely well on its training data but performs poorly when it receives new data. What issue does this behavior indicate?

1. Overfitting
2. Underfitting
3. Data encryption
4. Model deployment

Correct Answer: 1

Explanation

Overfitting occurs when a machine learning model learns the training data too closely, including patterns that do not generalize well to new examples. Such a model may achieve excellent performance on training data while producing significantly poorer results on unseen data. This can occur when a model is excessively complex relative to the available training information. Underfitting represents the opposite situation, where the model fails to capture important patterns even in the training data. Evaluating performance on unseen validation or test data helps identify generalization problems.

Question 118

An organization wants to create a virtual assistant that answers questions using company policies and also provides current information from an internal knowledge base. Which design approach can help keep responses grounded in that information?

1. Increasing image resolution
2. Using retrieval to provide relevant context
3. Removing all system instructions
4. Randomly changing the model temperature

Correct Answer: 2

Explanation

Retrieval can provide a generative AI model with relevant information from an organization’s knowledge base before the model generates a response. Supplying this context can help the application produce answers that are grounded in the organization’s available information. This approach is useful when knowledge changes over time or when the required information is specific to an organization. Increasing image resolution and changing temperature do not provide factual context, while removing system instructions can reduce control over application behavior. Grounded retrieval is therefore suitable for this virtual-assistant scenario.

Question 119

A company wants to compare a classification model’s ability to correctly identify positive cases while also considering false positives. Which metric combines precision and recall into a single measure?

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

Correct Answer: 3

Explanation

The F1 score combines precision and recall into a single metric by calculating their harmonic mean. It is useful when both false positives and false negatives matter and a balance between precision and recall is desired. Accuracy measures the proportion of all predictions that are correct, while mean squared error and mean absolute error are commonly associated with regression evaluation. The F1 score can therefore provide a useful summary of classification performance when simply measuring accuracy does not fully represent the business requirements.

Question 120

A developer needs an AI solution that can analyze the meaning and sentiment of written customer reviews. Which Azure service is most directly suited to this requirement?

1. Azure AI Vision
2. Azure AI Language
3. Azure AI Speech
4. Azure AI Document Intelligence

Correct Answer: 2

Explanation

Azure AI Language provides capabilities for analyzing and understanding written language. It can support scenarios such as sentiment analysis, key phrase extraction, entity recognition, language detection, summarization, and other text-processing tasks. Customer reviews are primarily text data, so Language is directly suited to analyzing their meaning and sentiment. Azure AI Vision focuses on visual information, Speech handles spoken language and audio, and Document Intelligence is designed primarily for extracting information from documents. Therefore, Azure AI Language is the appropriate service for analyzing written customer feedback.