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

 

Question 121

Which type of machine learning uses labeled examples to learn how input features relate to a known target value?

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

Correct Answer: 3

Explanation

Supervised learning uses training data that contains both input features and known target values, also called labels. The model learns the relationship between the inputs and the expected outputs so it can make predictions for new data. Classification and regression are common supervised learning tasks. In contrast, unsupervised learning works with data that does not have known labels, while reinforcement learning learns through interactions and rewards. Clustering is an unsupervised technique that groups similar observations without requiring predefined target categories.

Question 122

A retail company wants to predict whether a customer will cancel a subscription based on account activity, usage, and previous interactions. What type of machine learning problem is this?

1. Classification
2. Regression
3. Image segmentation
4. Clustering

Correct Answer: 1

Explanation

Predicting whether a customer will cancel a subscription involves selecting between categories, such as cancel or not cancel. This makes the problem a classification task. The model can learn from historical customer records where the final outcome is already known and then classify new customers based on their features. Regression is used when predicting continuous numerical values, while clustering groups unlabeled observations. Image segmentation is a computer vision task that separates regions of an image and does not match this customer-retention scenario.

Question 123

A company has thousands of product descriptions and wants to automatically identify the main topics discussed in each description without providing predefined categories. Which approach is most appropriate?

1. Regression
2. Supervised classification
3. Speech recognition
4. Unsupervised learning

Correct Answer: 4

Explanation

Unsupervised learning is appropriate when a solution needs to discover patterns or groupings in data without predefined labels. In this scenario, the product descriptions do not have existing categories, so an unsupervised technique can help identify naturally occurring groups or topics. Clustering is a common example of unsupervised learning. Supervised classification would require labeled examples, while regression predicts numerical values. Speech recognition processes spoken language rather than written product descriptions, so it is not appropriate for this requirement.

Question 124

An AI application must identify several different types of objects in a photograph and determine where each object appears. Which computer vision capability should be used?

1. Image classification
2. Object detection
3. OCR
4. Image tagging

Correct Answer: 2

Explanation

Object detection identifies objects within an image and determines their locations. For example, a photograph containing several vehicles and pedestrians can be analyzed to identify each object and locate it using bounding regions. Image classification generally assigns one or more categories to an entire image without providing individual object locations. OCR extracts written characters from visual content, while image tagging describes concepts present in an image. Because the requirement includes both identifying multiple objects and locating them, object detection is the suitable capability.

Question 125

A machine learning model uses customer age, annual income, and purchase frequency to make predictions. What are these values called in the model?

1. Features
2. Labels
3. Predictions
4. Endpoints

Correct Answer: 1

Explanation

Features are the input variables used by a machine learning model to identify patterns and generate predictions. In this example, age, annual income, and purchase frequency provide information that the model can use when making its prediction. Labels represent known target outcomes used during supervised training. Predictions are the outputs generated by the trained model, while an endpoint is a location through which a model or service can be accessed. Identifying features correctly is an important part of preparing data for machine learning.

Question 126

A healthcare organization trains a model using historical patient records where the desired outcome is already known. Why are these known outcomes important during supervised training?

1. They increase image resolution
2. They identify the training labels
3. They replace the model
4. They create an inference endpoint

Correct Answer: 2

Explanation

Known outcomes in supervised training serve as labels that tell the model what result is associated with each training example. The model compares its predictions with these known labels and adjusts its internal parameters during training to reduce errors. For example, historical records could indicate whether a particular outcome occurred, providing the target information needed for learning. Labels do not create endpoints or modify image resolution. They are essential because supervised algorithms require known target values to learn relationships between features and expected results.

Question 127

A machine learning team wants to test whether a trained model performs well on data it has never seen during training. Which dataset should they use?

1. Training dataset
2. Feature dataset
3. Test dataset
4. Label dataset

Correct Answer: 3

Explanation

A test dataset contains examples that were not used to train the model and is used to evaluate how well the trained model generalizes to unseen data. Testing with separate data provides a more realistic indication of how the model may perform when deployed. The training dataset is used to learn model parameters, while features and labels describe components of the data rather than specifically identifying an evaluation dataset. Keeping test data separate helps prevent overly optimistic evaluation results caused by testing on examples the model already encountered.

Question 128

A developer wants a machine learning model to generate predictions for thousands of historical records overnight rather than responding immediately to individual requests. Which inference pattern is most suitable?

1. Batch inference
2. Online inference
3. Interactive speech recognition
4. Real-time translation

Correct Answer: 1

Explanation

Batch inference is designed to process a collection of records together rather than generating a prediction for each request immediately. It is suitable when an organization needs to process large volumes of data on a scheduled basis, such as calculating predictions for thousands of historical transactions overnight. Online inference is more appropriate when applications require immediate responses to individual requests. Speech recognition and translation are specific AI capabilities rather than inference patterns for processing large datasets. Therefore, batch inference fits the overnight processing requirement.

Question 129

Which scenario is most suitable for online inference?

1. Processing ten years of archived records once a month
2. Generating a prediction immediately after a user submits an application
3. Recalculating all historical customer scores every weekend
4. Analyzing a large collection of stored files overnight

Correct Answer: 2

Explanation

Online inference is appropriate when an application needs a prediction or result shortly after receiving an individual request. For example, a loan application system might submit a customer’s information to a deployed model and immediately receive a prediction. Batch inference is better suited to scheduled processing of large collections of records. Historical analysis and overnight file processing generally do not require an immediate response for each individual record. Therefore, generating a prediction immediately after a user submits information is an example of online inference.

Question 130

A model predicts house prices, and the actual prices are known. Which evaluation approach is appropriate for measuring the average absolute difference between predictions and actual values?

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

Correct Answer: 4

Explanation

Mean absolute error, or MAE, measures the average absolute difference between predicted numerical values and their actual values. It is commonly used for evaluating regression models such as those that predict house prices. A lower MAE generally indicates that predictions are closer to the actual values. Precision and recall are classification metrics, while accuracy measures the proportion of correct predictions and is primarily used for classification scenarios. Since house prices are continuous numerical values, MAE provides an appropriate way to quantify prediction error.

Question 131

A model correctly identifies 90 out of 100 customer transactions as belonging to their correct classes. Which metric represents the proportion of correct predictions out of all predictions?

1. Accuracy
2. Recall
3. Precision
4. F1 score

Correct Answer: 1

Explanation

Accuracy measures the proportion of predictions that are correct across all evaluated examples. If a model correctly classifies 90 of 100 transactions, its accuracy is 90 percent. Accuracy considers both correctly predicted positive and negative cases in the overall result. Precision focuses on how many predicted positive cases are actually positive, while recall measures how many actual positive cases were identified. F1 score combines precision and recall. Therefore, the metric described by the proportion of all correct predictions is accuracy.

Question 132

An AI system is used to screen applications, but its performance is significantly different across relevant demographic groups. Which responsible AI principle should be investigated?

1. Transparency
2. Fairness
3. Reliability
4. Accountability

Correct Answer: 2

Explanation

Fairness requires AI systems to avoid unjustified differences in outcomes or performance across groups. If an application-screening model performs significantly differently for relevant demographic groups, the organization should investigate whether the training data, features, model behavior, or decision process contributes to unfair outcomes. Transparency focuses on making AI behavior understandable, reliability concerns consistent and dependable operation, and accountability concerns assigning responsibility for AI decisions. Investigating fairness is therefore important when an AI system produces materially different outcomes for different groups.

Question 133

A company wants users to understand why an AI system produced a particular recommendation. Which responsible AI principle is most directly related to this requirement?

1. Transparency
2. Inclusiveness
3. Privacy
4. Safety

Correct Answer: 1

Explanation

Transparency involves making information about an AI system’s behavior and decision-making understandable to relevant users and stakeholders. If users need to understand why a recommendation was produced, the organization should provide appropriate information about how the system operates and what factors influence its results. Inclusiveness focuses on designing systems that work effectively for diverse users, privacy concerns the protection of personal information, and safety focuses on preventing harmful or unsafe outcomes. Therefore, transparency is the principle most directly associated with explaining AI recommendations.

Question 134

A company wants its AI application to work effectively for people with different abilities and backgrounds. Which responsible AI principle is most relevant?

1. Accountability
2. Reliability
3. Inclusiveness
4. Transparency

Correct Answer: 3

Explanation

Inclusiveness means designing AI systems so they can provide useful experiences for people with different abilities, backgrounds, and circumstances. An organization should consider diverse user needs during design, development, testing, and deployment rather than assuming that one user group represents everyone. Accountability concerns responsibility for decisions and outcomes, reliability focuses on dependable operation, and transparency focuses on making system behavior understandable. When the main requirement is ensuring that an AI application can serve a diverse population effectively, inclusiveness is the relevant responsible AI principle.

Question 135

A generative AI application receives a long instruction that contains several separate tasks. The developer wants the model to follow a clear sequence when responding. Which prompt-design practice can help?

1. Provide clear and structured instructions
2. Remove all context from the prompt
3. Use unrelated examples
4. Replace the prompt with image data

Correct Answer: 1

Explanation

Clear and structured instructions can help a generative AI model understand the intended task and expected response format. When a prompt contains multiple requirements, organizing the instructions into logical sections or an explicit sequence can reduce ambiguity. Removing context may make the task harder, while unrelated examples can introduce confusion. Image data is not a replacement for textual instructions in a text-generation scenario. Good prompt design therefore involves communicating the task, context, constraints, and desired output clearly enough for the model to interpret the request.

Question 136

A generative AI application must provide consistent instructions about its role, behavior, and response boundaries across conversations. Which prompt component is designed for this purpose?

1. User message
2. System message
3. Search index
4. Test dataset

Correct Answer: 2

Explanation

A system message provides high-level instructions that guide how a generative AI model should behave during an interaction. It can define the assistant’s role, establish behavioral rules, specify response requirements, and set boundaries for the application. A user message contains the user’s particular request, while a search index stores information that can be retrieved by an application. A test dataset is used to evaluate machine learning performance. Therefore, a system message is the appropriate component for establishing consistent behavior and instructions.

Question 137

A company stores customer names, addresses, and identification numbers. Before using this information in an AI solution, which responsible AI concern is particularly important?

1. Privacy
2. Image resolution
3. Object detection
4. Model temperature

Correct Answer: 1

Explanation

Privacy is particularly important when an AI solution processes personally identifiable or otherwise sensitive information. Organizations should consider how personal data is collected, stored, accessed, processed, and protected throughout the AI system’s lifecycle. Appropriate controls can help reduce unauthorized access or inappropriate use of customer information. Image resolution and object detection relate to computer vision, while temperature is a generative AI setting that affects output randomness. Because the scenario involves names, addresses, and identification numbers, privacy should be a major consideration.

Question 138

An organization wants to extract handwritten and printed text from scanned forms so the content can be processed digitally. Which capability is required?

1. Object detection
2. Sentiment analysis
3. Optical character recognition
4. Speech synthesis

Correct Answer: 3

Explanation

Optical character recognition, or OCR, is used to detect and extract written or printed characters from images and scanned documents. It can convert visual text into machine-readable content that applications can store, search, or analyze. Object detection identifies objects and their locations, sentiment analysis evaluates opinions expressed in language, and speech synthesis converts text into spoken audio. For scanned forms containing handwritten or printed information, OCR is therefore the relevant capability for turning the visible text into digital data.

Question 139

A company wants to build a model that predicts the amount of electricity a building will consume tomorrow using historical consumption values and other relevant measurements. Which machine learning approach is most appropriate?

1. Classification
2. Clustering
3. Regression
4. Image segmentation

Correct Answer: 3

Explanation

Regression is used to predict continuous numerical values, making it appropriate for estimating future electricity consumption. The model can use historical consumption and other relevant measurements as input features to learn relationships with the target consumption value. Depending on the design, the problem can also involve time-series forecasting because the target is associated with a future point in time. Classification predicts categories, clustering groups observations without predefined labels, and image segmentation separates regions within images. The numerical prediction requirement makes regression a suitable approach.

Question 140

A team wants to ensure that an AI model continues performing correctly after deployment as real-world data changes over time. What activity should the team perform?

1. Remove the test dataset
2. Monitor the model and its data
3. Disable model evaluation
4. Increase every input value

Correct Answer: 2

Explanation

Monitoring a deployed AI solution helps the team identify changes in data, model behavior, performance, and operational conditions over time. Real-world data may change from the information used during training, potentially affecting prediction quality. Regular monitoring can help detect performance degradation and indicate when investigation, retraining, or other corrective action may be necessary. Removing test data or disabling evaluation would make it harder to identify problems, while changing input values does not provide a monitoring strategy. Continuous monitoring is therefore important for maintaining reliable AI solutions.