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

 

Question 261

A bank wants to predict whether a loan applicant will default based on income, credit history, and existing debt. Which machine learning task is appropriate?

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

Correct Answer: 3

Explanation

Classification is appropriate when a model needs to assign an input to one of several predefined categories. In this scenario, the model can predict whether an applicant will default or not default, creating a binary classification problem. Income, credit history, and existing debt can be used as features during training. Regression would be appropriate if the target were a numerical amount, while clustering discovers groups without predefined labels. OCR is unrelated because the primary requirement is making a categorical prediction from applicant information.

Question 262

A company wants to discover naturally occurring groups among products based on purchasing patterns without providing predefined product categories. Which technique should be used?

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

Correct Answer: 2

Explanation

Clustering is an unsupervised learning technique that discovers naturally occurring groups within data without requiring predefined categories. A retailer can provide purchasing patterns, product attributes, or customer interactions and allow a clustering algorithm to identify products that behave similarly. Classification requires known categories for training, while regression predicts numerical values. Speech recognition processes spoken language and does not address product grouping. Therefore, clustering is the appropriate technique when the goal is to discover meaningful groups without supplying predefined labels.

Question 263

A model predicts the number of support tickets a company will receive each day. The output can be any reasonable numerical value. Which type of prediction is this?

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

Correct Answer: 3

Explanation

Regression is used to predict continuous numerical values. In this example, the model estimates the number of support tickets expected on a particular day, making the target a numerical quantity rather than a predefined category. Features such as historical ticket counts, day of the week, product releases, or seasonal trends could be used to train the model. Classification predicts categories, entity recognition analyzes text, and clustering discovers groups. Because the required output is a numerical prediction, regression is the appropriate machine learning task.

Question 264

A mobile application allows users to take pictures of street signs and automatically convert the visible words into editable text. Which capability is required?

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

Correct Answer: 2

Explanation

OCR, or optical character recognition, extracts written characters and words from images and converts them into machine-readable text. A mobile application can use OCR to analyze photographs of street signs and provide editable text to the user. Object detection identifies and locates objects, but it does not primarily convert printed words into editable text. Sentiment analysis evaluates opinions in language, while speech synthesis generates spoken audio from text. Therefore, OCR is the appropriate computer vision capability for extracting the words displayed on street signs.

Question 265

A customer service chatbot must understand that “I cannot access my account” and “I’m locked out” describe a similar request. Which AI workload is most relevant?

1. Natural language processing
2. Image classification
3. Object detection
4. Regression

Correct Answer: 1

Explanation

Natural language processing enables applications to analyze and interpret human language. A customer service chatbot can use NLP capabilities to understand that different phrases may communicate similar meanings or intents. This allows the application to route users to an appropriate response even when their wording differs. Image classification categorizes images, object detection locates objects within images, and regression predicts numerical values. Because the requirement involves understanding the meaning and intent of different text expressions, natural language processing is the relevant AI workload.

Question 266

An AI assistant should always answer customer questions using a formal and concise style defined by the application developer. Which approach can establish this behavior?

1. Image labeling
2. A system message
3. Batch inference
4. OCR

Correct Answer: 2

Explanation

A system message can establish high-level instructions that influence how a generative AI assistant behaves. Developers can use it to specify requirements such as tone, response format, role, or behavioral constraints. The assistant can then use those instructions when processing individual user requests. Image labeling and OCR address visual information, while batch inference concerns how model predictions are processed. Since the requirement is to establish consistent response behavior across interactions, a system message is the most appropriate approach.

Question 267

A company wants an AI model to recognize defective products from photographs taken on a manufacturing line. Which AI workload is most directly involved?

1. Computer vision
2. Speech synthesis
3. Natural language translation
4. Time-series forecasting

Correct Answer: 1

Explanation

Computer vision enables AI systems to analyze and interpret visual information contained in images or video. A manufacturing application can use computer vision to inspect product photographs and identify visual characteristics associated with defects. Depending on the requirement, image classification, object detection, or other vision capabilities may be used. Speech synthesis produces spoken audio, translation converts language between languages, and time-series forecasting predicts future values. Because the input consists of photographs and the system must analyze their visual content, computer vision is the relevant AI workload.

Question 268

An organization wants to extract invoice number, date, vendor name, and total amount from many different invoice layouts. Which capability is most suitable when the organization needs a model tailored to its own documents?

1. Custom document extraction
2. Speech recognition
3. Sentiment analysis
4. Image captioning

Correct Answer: 1

Explanation

Custom document extraction can be used when an organization needs to extract specific fields from documents whose layouts or content requirements are not adequately addressed by a standard prebuilt model. The organization can train a custom model using representative documents and identify the fields it needs, such as invoice number, date, vendor name, and total amount. Speech recognition handles audio, sentiment analysis evaluates text opinions, and image captioning generates descriptions of images. Therefore, a custom document extraction approach is appropriate for organization-specific invoice processing requirements.

Question 269

A model is evaluated using a confusion matrix. Which information can the matrix help the team understand?

1. The model’s training cost only
2. The number of tokens in a prompt
3. Correct and incorrect classification outcomes
4. The physical size of an image

Correct Answer: 3

Explanation

A confusion matrix summarizes classification results by showing categories such as true positives, true negatives, false positives, and false negatives. These values help teams understand how a classification model performs and can be used to calculate metrics such as accuracy, precision, recall, and F1 score. A confusion matrix does not describe training cost, prompt token counts, or image dimensions. By examining the different types of correct and incorrect predictions, a development team can gain more detailed insight into the behavior of a classification model.

Question 270

A model identifies 90 emails as spam, and 72 of those emails are actually spam. Which metric measures the proportion of predicted spam messages that were correct?

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

Correct Answer: 4

Explanation

Precision measures the proportion of predicted positive cases that are actually positive. In this scenario, the model predicted 90 emails as spam, and 72 of those predictions were correct. Therefore, precision is calculated from the number of true positives divided by the total number of predicted positives. Recall would instead measure how many of all actual spam messages were detected. Accuracy considers all correct predictions, while mean absolute error is mainly used for numerical predictions. Thus, precision is the metric that matches the stated requirement.

Question 271

An AI application must identify the language used in a customer message before selecting an appropriate translation service. Which capability can perform this initial task?

1. Language detection
2. Object detection
3. Regression
4. Text-to-speech

Correct Answer: 1

Explanation

Language detection identifies the language in which a piece of text is written. This capability can help an application determine whether a customer message is written in English, Urdu, Spanish, French, or another supported language before selecting an appropriate processing or translation workflow. Object detection works with visual content, regression predicts numerical values, and text-to-speech converts written content into spoken audio. Since the application first needs to determine the language of an incoming written message, language detection is the appropriate capability.

Question 272

A company wants its AI application to answer questions from an internal policy library while keeping the source material outside the model’s original training data. Which approach is appropriate?

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

Correct Answer: 1

Explanation

Retrieval-augmented generation allows an application to retrieve relevant information from an external source and provide that information to a generative model as context. An internal policy library can therefore remain an external knowledge source while the application retrieves the appropriate passages when users ask questions. This approach can help generate responses based on organization-specific information without requiring all of that information to be part of the model’s original training data. Image classification, speech synthesis, and regression address different workloads and do not provide this retrieval-based approach.

Question 273

Which data type is most appropriate for storing information such as a customer’s age, account balance, and number of purchases?

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

Correct Answer: 2

Explanation

Numerical data represents values that can be expressed as numbers and used for calculations or statistical analysis. Customer age, account balance, and number of purchases are examples of numerical information. Such values can be used as features in machine learning models and may require appropriate preprocessing depending on the algorithm. Audio, image, and video are other data types designed for different forms of information. Because all three examples in the scenario are numeric measurements or counts, numerical data is the appropriate data type.

Question 274

A generative AI application is producing overly predictable answers when developers want more varied responses for brainstorming. Which change may increase output variability?

1. Lowering the temperature substantially
2. Increasing the temperature
3. Removing all prompts
4. Converting text into images

Correct Answer: 2

Explanation

Increasing the temperature can make generative AI responses more varied by allowing the model to consider a broader range of possible token selections. This can be useful for creative brainstorming where diversity of ideas is desirable. Lower temperatures generally encourage more predictable and deterministic outputs. Removing prompts would not provide controlled behavior, while converting text into images is unrelated to response variability. Temperature should still be selected carefully because excessive randomness may reduce consistency or usefulness. Therefore, increasing temperature can be appropriate when greater output variation is desired.

Question 275

A company evaluates an AI system with data that closely reflects the people, situations, and conditions expected after deployment. What quality of the evaluation data is being demonstrated?

1. Representativeness
2. Compression
3. Encryption
4. Resolution

Correct Answer: 1

Explanation

Representativeness means that data adequately reflects the population, conditions, and scenarios in which an AI system is expected to operate. Evaluation data with representative characteristics can provide a more realistic indication of how the system may perform after deployment. If important groups or situations are missing, evaluation results may not accurately reflect real-world behavior. Compression and encryption concern data handling, while resolution generally relates to image or video quality. Therefore, when evaluation data resembles the expected real-world operating environment, it demonstrates representativeness.

Question 276

A company wants to generate a short description of what is visible in a photograph. Which computer vision capability is most suitable?

1. Image captioning
2. Speech recognition
3. Key phrase extraction
4. Regression

Correct Answer: 1

Explanation

Image captioning generates a natural-language description of the content visible in an image. For example, an application could analyze a photograph and produce a description such as a person standing beside a vehicle. This differs from object detection, which identifies and locates individual objects, and image classification, which assigns categories to an image. Speech recognition works with spoken audio, key phrase extraction analyzes written language, and regression predicts numerical values. Since the requirement is to generate a textual description of an image, image captioning is appropriate.

Question 277

An AI application is expected to continue working correctly even when normal operating conditions vary. Which responsible AI principle is most closely associated with this requirement?

1. Fairness
2. Reliability and safety
3. Transparency
4. Inclusiveness

Correct Answer: 2

Explanation

Reliability and safety focus on ensuring that an AI system performs consistently and safely under expected operating conditions, including situations where conditions may vary. Developers should consider testing, monitoring, failure handling, and appropriate safeguards to help maintain dependable system behavior. Fairness addresses equitable treatment, transparency concerns understandable information about AI behavior, and inclusiveness focuses on supporting people with different abilities and circumstances. Since the scenario emphasizes continued correct and safe operation under changing normal conditions, reliability and safety are the most relevant responsible AI considerations.

Question 278

A model’s training accuracy is extremely high, but its performance on new evaluation data is significantly lower. Which problem does this pattern suggest?

1. Underfitting
2. Overfitting
3. Clustering
4. Language detection

Correct Answer: 2

Explanation

Overfitting occurs when a model learns the training data too closely, including patterns that do not generalize well to unseen data. A common sign is very strong performance on training data combined with substantially weaker performance on evaluation or test data. The model may effectively memorize aspects of the training examples instead of learning patterns that apply more broadly. Underfitting generally results in poor performance even on training data. Clustering and language detection are unrelated concepts. Therefore, the described difference between training and unseen-data performance suggests overfitting.

Question 279

An AI solution needs to process spoken English and provide the same message in spoken French. Which sequence of capabilities can support this requirement?

1. OCR followed by sentiment analysis
2. Speech-to-text, translation, then text-to-speech
3. Object detection followed by classification
4. Regression followed by clustering

Correct Answer: 2

Explanation

Speech-to-text can first convert the spoken English input into written text. A translation capability can then convert the English text into French. Finally, text-to-speech can synthesize the translated French text as spoken audio. This sequence connects speech recognition, language translation, and speech synthesis into a complete spoken-language workflow. OCR is intended for text contained in images, while sentiment analysis evaluates opinions. Object detection, classification, regression, and clustering do not provide the required end-to-end speech translation process.

Question 280

A developer wants an AI assistant to provide answers based on retrieved documents while clearly distinguishing those documents from the model’s general knowledge. What should the application provide to the model?

1. Relevant retrieved context
2. Higher image resolution
3. A larger confusion matrix
4. Additional audio volume

Correct Answer: 1

Explanation

Providing relevant retrieved context allows a generative AI model to base its response on information supplied by an external source. This approach is commonly used with retrieval-augmented generation and grounding, where application logic retrieves useful documents or passages and presents them to the model as context. The retrieved material can help the model produce responses that are more closely tied to the organization’s information. Image resolution, confusion matrices, and audio volume do not provide textual knowledge for grounding a generative response.