{"id":16125,"date":"2026-09-18T11:22:00","date_gmt":"2026-09-18T11:22:00","guid":{"rendered":"https:\/\/www.examlabs.com\/certification\/?p=16125"},"modified":"2026-09-18T11:22:00","modified_gmt":"2026-09-18T11:22:00","slug":"microsoft-ai-901-practice-test-questions-and-exam-dumps-part-19-q361-380","status":"publish","type":"post","link":"https:\/\/www.examlabs.com\/certification\/microsoft-ai-901-practice-test-questions-and-exam-dumps-part-19-q361-380\/","title":{"rendered":"Microsoft AI-901 Practice Test Questions and Exam Dumps Part 19 Q361-380"},"content":{"rendered":"

<\/h1>\n

View Full <\/b>Microsoft AI-901 Exam Dumps<\/b><\/a> and Practice Test Dumps<\/b><\/p>\n

Question 361. Which Azure service provides a centralized environment for developing and evaluating generative AI applications?<\/b><\/h3>\n
    \n
  1. <\/b> Azure AI Foundry<\/span><\/li>\n
  2. <\/b> Azure Virtual Network<\/span><\/li>\n
  3. <\/b> Azure DNS<\/span><\/li>\n
  4. <\/b> Azure Storage Explorer<\/span><\/li>\n<\/ol>\n

    Correct Answer: 1. Azure AI Foundry<\/b><\/p>\n

    Explanation:<\/b><\/p>\n

    Azure AI Foundry provides an integrated environment for developing, testing, evaluating, and managing generative AI applications. It can bring together models, prompts, evaluation capabilities, application components, and monitoring workflows in one development experience. Teams can use it to experiment with different models, assess response quality, and prepare applications for deployment. The other options serve infrastructure or storage-related purposes rather than providing a dedicated AI development and evaluation environment. For AI solution development, a centralized workspace such as Azure AI Foundry can help organize experimentation and improve consistency throughout the application lifecycle.<\/span><\/p>\n

    Question 362. A developer wants an AI model to answer questions using only information from a company knowledge base. What technique should be used?<\/b><\/h3>\n
      \n
    1. <\/b> Random sampling<\/span><\/li>\n
    2. <\/b> Grounding<\/span><\/li>\n
    3. <\/b> Data normalization<\/span><\/li>\n
    4. <\/b> Image segmentation<\/span><\/li>\n<\/ol>\n

      Correct Answer: 2. Grounding<\/b><\/p>\n

      Explanation:<\/b><\/p>\n

      Grounding connects an AI model’s responses to trusted external information instead of relying entirely on information learned during model training. In a company knowledge-base scenario, retrieved documents can provide the factual context needed to answer a user’s question. This can reduce unsupported or fabricated responses and help the application provide information that reflects the organization’s current content. Grounding is especially useful for enterprise applications where answers should be based on internal policies, procedures, product documentation, or other controlled sources. The retrieved information should also be properly authorized so users do not receive content they are not permitted to access.<\/span><\/p>\n

      Question 363. Which search approach combines keyword matching with vector similarity to improve document retrieval?<\/b><\/h3>\n
        \n
      1. <\/b> Image classification<\/span><\/li>\n
      2. <\/b> Speech recognition<\/span><\/li>\n
      3. <\/b> Hybrid search<\/span><\/li>\n
      4. <\/b> Regression<\/span><\/li>\n<\/ol>\n

        Correct Answer: 3. Hybrid search<\/b><\/p>\n

        Explanation:<\/b><\/p>\n

        Hybrid search combines traditional keyword-based retrieval with vector-based similarity search. Keyword search is useful when exact terms, names, identifiers, or specific phrases are important, while vector search can identify content that is semantically similar even when different words are used. Combining the two approaches can improve retrieval quality for enterprise applications and retrieval-augmented generation systems. Azure AI Search supports search capabilities that can combine lexical and vector approaches. This is particularly valuable when users may express the same concept in different ways or when both exact matching and semantic relevance are important for finding useful documents.<\/span><\/p>\n

        Question 364. What is the primary purpose of embeddings in an AI search solution?<\/b><\/h3>\n
          \n
        1. <\/b> To convert text into numerical vectors representing meaning<\/span><\/li>\n
        2. <\/b> To encrypt database passwords<\/span><\/li>\n
        3. <\/b> To compress images into smaller files<\/span><\/li>\n
        4. <\/b> To schedule model training jobs<\/span><\/li>\n<\/ol>\n

          Correct Answer: 1. To convert text into numerical vectors representing meaning<\/b><\/p>\n

          Explanation:<\/b><\/p>\n

          Embeddings represent information such as text as numerical vectors in a mathematical space. Texts with related meanings can have vectors that are relatively close to one another, allowing applications to perform semantic similarity searches. This is different from traditional keyword matching, which primarily looks for literal terms. Embeddings are commonly used with vector search and retrieval-augmented generation systems to locate relevant content based on meaning. For example, a user may ask a question using words that do not exactly appear in a document, but an embedding-based search can still retrieve the document if its content expresses a closely related concept.<\/span><\/p>\n

          Question 365. Which practice helps protect sensitive information when building an AI application?<\/b><\/h3>\n
            \n
          1. <\/b> Collecting every available user attribute<\/span><\/li>\n
          2. <\/b> Disabling access controls<\/span><\/li>\n
          3. <\/b> Storing credentials in prompts<\/span><\/li>\n
          4. <\/b> Applying data minimization<\/span><\/li>\n<\/ol>\n

            Correct Answer: 4. Applying data minimization<\/b><\/p>\n

            Explanation:<\/b><\/p>\n

            Data minimization means collecting, processing, and retaining only the information necessary for the intended purpose. This practice can reduce the amount of sensitive information exposed to an AI system and limit the potential impact of unauthorized access or accidental disclosure. Developers should avoid sending unnecessary personal or confidential information to models, prompts, logs, or external services. Access controls, encryption, secure credential management, and appropriate retention policies can complement data minimization. Together, these practices support privacy and security throughout an AI application’s lifecycle and help organizations reduce unnecessary exposure of sensitive data.<\/span><\/p>\n

            Question 366. Which metric measures the proportion of predicted positive results that are actually positive?<\/b><\/h3>\n
              \n
            1. <\/b> Recall<\/span><\/li>\n
            2. <\/b> Precision<\/span><\/li>\n
            3. <\/b> Mean squared error<\/span><\/li>\n
            4. <\/b> Accuracy<\/span><\/li>\n<\/ol>\n

              Correct Answer: 2. Precision<\/b><\/p>\n

              Explanation:<\/b><\/p>\n

              Precision measures how many of the instances predicted as positive are actually positive. It is calculated as true positives divided by the total number of predicted positives, which includes true positives and false positives. Precision is particularly important when false positive predictions can create significant costs or operational problems. For example, if an automated system flags transactions as potentially fraudulent, high precision means that a larger proportion of flagged transactions are genuinely suspicious. Precision differs from recall, which measures how many of the actual positive cases were successfully identified by the model.<\/span><\/p>\n

              Question 367. Which metric is especially useful when missing positive cases is more costly than generating some false positives?<\/b><\/h3>\n
                \n
              1. <\/b> Recall<\/span><\/li>\n
              2. <\/b> Precision<\/span><\/li>\n
              3. <\/b> Mean absolute error<\/span><\/li>\n
              4. <\/b> R-squared<\/span><\/li>\n<\/ol>\n

                Correct Answer: 1. Recall<\/b><\/p>\n

                Explanation:<\/b><\/p>\n

                Recall measures the proportion of actual positive cases that a model correctly identifies. It is calculated using true positives divided by the total number of actual positives, including both true positives and false negatives. Recall is especially important when failing to detect a positive case can have serious consequences. For example, in a screening application, missing a relevant case may be more problematic than reviewing some additional false alarms. Although precision and recall should often be considered together, the appropriate balance depends on the application’s business requirements, risk tolerance, and consequences associated with different types of errors.<\/span><\/p>\n

                Question 368. What is the main purpose of a validation dataset during machine learning development?<\/b><\/h3>\n
                  \n
                1. <\/b> To permanently store production predictions<\/span><\/li>\n
                2. <\/b> To replace the training dataset<\/span><\/li>\n
                3. <\/b> To tune and compare model configurations during development<\/span><\/li>\n
                4. <\/b> To encrypt model files<\/span><\/li>\n<\/ol>\n

                  Correct Answer: 3. To tune and compare model configurations during development<\/b><\/p>\n

                  Explanation:<\/b><\/p>\n

                  A validation dataset is used during model development to evaluate different configurations without using the final test dataset for repeated tuning. Developers can use validation results to compare models, adjust hyperparameters, select features, or identify potential overfitting. After development is complete, a separate test dataset can provide a more independent estimate of how the selected model performs on unseen data. Keeping training, validation, and test data appropriately separated helps produce more reliable evaluation results. Using the test dataset repeatedly during development can cause information leakage and make the final performance measurement less trustworthy.<\/span><\/p>\n

                  Question 369. Which situation is an example of overfitting?<\/b><\/h3>\n
                    \n
                  1. <\/b> A model performs poorly on both training and validation data<\/span><\/li>\n
                  2. <\/b> A model performs similarly on training and unseen data<\/span><\/li>\n
                  3. <\/b> A model has too few features to learn a pattern<\/span><\/li>\n
                  4. <\/b> A model performs extremely well on training data but poorly on unseen data<\/span><\/li>\n<\/ol>\n

                    Correct Answer: 4. A model performs extremely well on training data but poorly on unseen data<\/b><\/p>\n

                    Explanation:<\/b><\/p>\n

                    Overfitting occurs when a model learns the training data too closely, including patterns or noise that do not generalize to new examples. As a result, the model can achieve very strong performance on its training dataset while performing substantially worse on validation or test data. This indicates that the model has not learned a generalizable representation of the underlying problem. Techniques such as regularization, simplifying the model, using more representative training data, cross-validation, or adjusting hyperparameters can help address overfitting. The appropriate solution depends on the model, dataset, and application requirements.<\/span><\/p>\n

                    Question 370. What is the primary purpose of a test dataset?<\/b><\/h3>\n
                      \n
                    1. <\/b> To provide a final evaluation of model performance on unseen data<\/span><\/li>\n
                    2. <\/b> To train the model’s parameters<\/span><\/li>\n
                    3. <\/b> To store application credentials<\/span><\/li>\n
                    4. <\/b> To define the model’s system prompt<\/span><\/li>\n<\/ol>\n

                      Correct Answer: 1. To provide a final evaluation of model performance on unseen data<\/b><\/p>\n

                      Explanation:<\/b><\/p>\n

                      A test dataset is reserved for evaluating a trained and selected model on data that was not used to fit its parameters or repeatedly tune its configuration. This provides an estimate of how the model may perform on unseen examples in the intended environment. Separating the test dataset from training and validation data helps reduce evaluation bias. If developers repeatedly adjust the model based on test results, the test set can effectively become part of the development process and lose its independence. A properly managed test dataset therefore supports more credible final performance measurement.<\/span><\/p>\n

                      Question 371. Which Azure Machine Learning capability helps automate the selection and tuning of machine learning models?<\/b><\/h3>\n
                        \n
                      1. <\/b> Azure DNS<\/span><\/li>\n
                      2. <\/b> Automated ML<\/span><\/li>\n
                      3. <\/b> Azure Firewall<\/span><\/li>\n
                      4. <\/b> Azure Storage Explorer<\/span><\/li>\n<\/ol>\n

                        Correct Answer: 2. Automated ML<\/b><\/p>\n

                        Explanation:<\/b><\/p>\n

                        Automated ML, or Automated Machine Learning, can automate portions of the machine learning model development process. Depending on the task and configuration, it can evaluate algorithms, preprocessing approaches, and hyperparameter combinations to identify promising models. This can help developers establish a baseline or accelerate experimentation without manually testing every possible configuration. Automated ML does not remove the need for understanding the business problem, preparing appropriate data, evaluating results, and validating the model in the intended context. Human review remains important because the automatically selected model still needs to meet functional, performance, fairness, and operational requirements.<\/span><\/p>\n

                        Question 372. What should an organization use to ensure an AI application does not expose documents to unauthorized users?<\/b><\/h3>\n
                          \n
                        1. <\/b> Larger prompts<\/span><\/li>\n
                        2. <\/b> More training epochs<\/span><\/li>\n
                        3. <\/b> Access controls and authorization checks<\/span><\/li>\n
                        4. <\/b> Higher image resolution<\/span><\/li>\n<\/ol>\n

                          Correct Answer: 3. Access controls and authorization checks<\/b><\/p>\n

                          Explanation:<\/b><\/p>\n

                          Access controls and authorization checks are essential when an AI application retrieves information from private organizational sources. Retrieval should respect the permissions associated with the user and the underlying documents. Without proper authorization, a system could retrieve sensitive information and present it to someone who should not have access, even if the generated response itself appears accurate. In retrieval-augmented generation systems, security therefore needs to be integrated into the retrieval process rather than added only after text generation. Identity management, document-level permissions, filtering, and secure data handling can work together to reduce unauthorized disclosure.<\/span><\/p>\n

                          Question 373. What does data drift describe in a deployed machine learning system?<\/b><\/h3>\n
                            \n
                          1. <\/b> A change in the distribution of incoming input data over time<\/span><\/li>\n
                          2. <\/b> A change in the model’s source code formatting<\/span><\/li>\n
                          3. <\/b> A failure to save model files<\/span><\/li>\n
                          4. <\/b> A reduction in storage capacity<\/span><\/li>\n<\/ol>\n

                            Correct Answer: 1. A change in the distribution of incoming input data over time<\/b><\/p>\n

                            Explanation:<\/b><\/p>\n

                            Data drift occurs when the statistical characteristics or distribution of input data changes over time compared with the data used to train the model. For example, customer behavior, geographic patterns, product usage, or seasonal conditions may change after deployment. These changes can affect model performance even if the underlying relationship between inputs and the target has not changed. Monitoring input distributions can help identify possible drift early. When meaningful changes are detected, teams may investigate the cause, evaluate model performance, and determine whether retraining or other updates are appropriate.<\/span><\/p>\n

                            Question 374. What does concept drift refer to?<\/b><\/h3>\n
                              \n
                            1. <\/b> A change in file storage location<\/span><\/li>\n
                            2. <\/b> A change in the relationship between input features and the target<\/span><\/li>\n
                            3. <\/b> A change in image resolution<\/span><\/li>\n
                            4. <\/b> A change in user interface design<\/span><\/li>\n<\/ol>\n

                              Correct Answer: 2. A change in the relationship between input features and the target<\/b><\/p>\n

                              Explanation:<\/b><\/p>\n

                              Concept drift occurs when the relationship between input variables and the target outcome changes over time. A model may have been trained correctly on historical data, but the patterns it learned may become less accurate as real-world conditions change. For example, customer behavior may change so that the same input characteristics no longer predict purchasing behavior in the same way. Concept drift differs from ordinary data drift, which concerns changes in the distribution of input data. Monitoring both types of change is important because either can contribute to declining model performance after deployment.<\/span><\/p>\n

                              Question 375. Which technique can improve retrieval by filtering documents according to attributes such as department or document type?<\/b><\/h3>\n
                                \n
                              1. <\/b> Random initialization<\/span><\/li>\n
                              2. <\/b> Image augmentation<\/span><\/li>\n
                              3. <\/b> Metadata filtering<\/span><\/li>\n
                              4. <\/b> Speech synthesis<\/span><\/li>\n<\/ol>\n

                                Correct Answer: 3. Metadata filtering<\/b><\/p>\n

                                Explanation:<\/b><\/p>\n

                                Metadata filtering allows a search system to restrict retrieval results according to attributes associated with documents. Examples include department, document type, language, date, region, security classification, or product category. In enterprise search applications, metadata filtering can improve relevance by narrowing the search space before or during retrieval. It can also support security requirements when metadata is appropriately tied to authorization rules. Metadata filtering is particularly useful in retrieval-augmented generation because the quality and appropriateness of retrieved documents directly affect the context supplied to the generative model.<\/span><\/p>\n

                                Question 376. Why might an organization choose a smaller generative AI model for a simple application?<\/b><\/h3>\n
                                  \n
                                1. <\/b> Smaller models always produce more accurate answers<\/span><\/li>\n
                                2. <\/b> Smaller models eliminate all security risks<\/span><\/li>\n
                                3. <\/b> Smaller models never require evaluation<\/span><\/li>\n
                                4. <\/b> Smaller models may reduce cost and latency when their capabilities are sufficient<\/span><\/li>\n<\/ol>\n

                                  Correct Answer: 4. Smaller models may reduce cost and latency when their capabilities are sufficient<\/b><\/p>\n

                                  Explanation:<\/b><\/p>\n

                                  A smaller generative AI model can sometimes provide an appropriate balance between capability, response quality, latency, and operational cost for a relatively simple task. Larger models may provide stronger performance for complex reasoning or broader tasks, but they can require more computational resources and may introduce higher costs or latency. Model selection should therefore consider the application’s requirements rather than assuming that the largest model is always appropriate. Organizations should evaluate candidate models against representative workloads, quality requirements, safety expectations, latency targets, and cost constraints before deployment.<\/span><\/p>\n

                                  Question 377. What is a key purpose of Azure AI Content Safety?<\/b><\/h3>\n
                                    \n
                                  1. <\/b> To create database indexes<\/span><\/li>\n
                                  2. <\/b> To help detect and manage potentially harmful content<\/span><\/li>\n
                                  3. <\/b> To train regression models<\/span><\/li>\n
                                  4. <\/b> To convert speech into audio files<\/span><\/li>\n<\/ol>\n

                                    Correct Answer: 2. To help detect and manage potentially harmful content<\/b><\/p>\n

                                    Explanation:<\/b><\/p>\n

                                    Azure AI Content Safety provides capabilities that can help applications identify potentially harmful content in user-generated or AI-generated material. Depending on the scenario, organizations can use content safety capabilities to analyze categories of harmful text or images and apply application-specific safeguards. Content filtering is an important component of responsible AI deployment because generative systems can sometimes produce or receive inappropriate content. However, automated safety detection should be considered as one part of a broader safety strategy that can include policy definitions, human review, access controls, testing, monitoring, and incident-response procedures.<\/span><\/p>\n

                                    Question 378. Which principle requires an organization to identify who is responsible for an AI system’s decisions and outcomes?<\/b><\/h3>\n
                                      \n
                                    1. <\/b> Accountability<\/span><\/li>\n
                                    2. <\/b> Compression<\/span><\/li>\n
                                    3. <\/b> Tokenization<\/span><\/li>\n
                                    4. <\/b> Vectorization<\/span><\/li>\n<\/ol>\n

                                      Correct Answer: 1. Accountability<\/b><\/p>\n

                                      Explanation:<\/b><\/p>\n

                                      Accountability means that appropriate people or organizations remain responsible for the development, deployment, operation, and outcomes of an AI system. An AI application should not be treated as independently responsible for important decisions. Organizations should establish clear ownership, governance procedures, monitoring processes, escalation paths, and documentation. Accountability is particularly important for systems that may affect users, customers, employees, or other stakeholders. Clear responsibilities help organizations investigate problems, respond to incidents, evaluate system performance, and make appropriate changes when the system does not behave as intended.<\/span><\/p>\n

                                      Question 379. Why should evaluation datasets contain representative examples of the intended users and scenarios?<\/b><\/h3>\n
                                        \n
                                      1. <\/b> To make the dataset as small as possible<\/span><\/li>\n
                                      2. <\/b> To avoid measuring model performance<\/span><\/li>\n
                                      3. <\/b> To better assess how the model performs in realistic conditions<\/span><\/li>\n
                                      4. <\/b> To eliminate the need for monitoring<\/span><\/li>\n<\/ol>\n

                                        Correct Answer: 3. To better assess how the model performs in realistic conditions<\/b><\/p>\n

                                        Explanation:<\/b><\/p>\n

                                        Representative evaluation data helps determine whether an AI system performs appropriately across the users, inputs, conditions, and scenarios it is expected to encounter after deployment. If an evaluation dataset contains only a narrow subset of possible users or situations, performance measurements may not reveal important weaknesses. Representative data can also help identify differences in performance across relevant groups and edge cases. Evaluation should therefore reflect the intended use context as closely as practical while respecting privacy, security, and data-governance requirements. The results can then guide improvements before and after deployment.<\/span><\/p>\n

                                        Question 380. What is an important activity after deploying an AI model to production?<\/b><\/h3>\n
                                          \n
                                        1. <\/b> Stop evaluating the model permanently<\/span><\/li>\n
                                        2. <\/b> Monitor performance, safety, and data changes<\/span><\/li>\n
                                        3. <\/b> Delete all evaluation records<\/span><\/li>\n
                                        4. <\/b> Disable access controls<\/span><\/li>\n<\/ol>\n

                                          Correct Answer: 2. Monitor performance, safety, and data changes<\/b><\/p>\n

                                          Explanation:<\/b><\/p>\n

                                          Deployment is not the end of the AI model lifecycle. Production systems should be monitored to identify changes in performance, input data, safety behavior, latency, errors, and other operational characteristics. Monitoring can reveal data drift, concept drift, unexpected usage patterns, or newly emerging failure cases. Teams can use these observations to investigate problems and determine whether the model, prompts, retrieval sources, safeguards, or other application components need to be updated. Continuous monitoring supports responsible operation because real-world conditions can change after deployment, even when the original evaluation results were satisfactory.<\/span><\/p>\n

                                           <\/p>\n","protected":false},"excerpt":{"rendered":"

                                          View Full Microsoft AI-901 Exam Dumps and Practice Test Dumps Question 361. Which Azure service provides a centralized environment for developing and evaluating generative AI applications? Azure AI Foundry Azure Virtual Network Azure DNS Azure Storage Explorer Correct Answer: 1. Azure AI Foundry Explanation: Azure AI Foundry provides an integrated environment for developing, testing, evaluating, […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[1648,1647],"tags":[],"_links":{"self":[{"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/posts\/16125"}],"collection":[{"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/comments?post=16125"}],"version-history":[{"count":1,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/posts\/16125\/revisions"}],"predecessor-version":[{"id":16128,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/posts\/16125\/revisions\/16128"}],"wp:attachment":[{"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/media?parent=16125"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/categories?post=16125"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/tags?post=16125"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}