McAfee Secure

Exam Code: AI-901

Exam Name: Microsoft Azure AI Fundamentals

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Azure AI Fundamentals

Microsoft AI-901 Questions & Answers

Study with Up-To-Date REAL Exam Questions and Answers from the ACTUAL Test

166 Questions & Answers with Testing Engine

"Microsoft Azure AI Fundamentals Exam", also known as AI-901 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-901 Exam Engine. Your AI-901 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-901 Practice Test 23

Question 220

An Azure AI Fundamentals solution has this requirement: A team improves a neural model by adding more epochs but the validation error worsens. Which implementation most directly addresses it?

  1. Use an image-classification model with class probabilities
  2. Confirm the model deployment identifier and endpoint resource match
  3. Use a held-out evaluation and investigate overfitting or data leakage
  4. Stop or regularize training based on held-out validation behavior

Explanation

Correct Answer

D:

More optimization on training data can worsen generalization. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Classification assigns image-wide categories rather than bounding-box coordinates. That is not sufficient here: More optimization on training data can worsen generalization.

B:

Applications call a deployed model resource rather than a generic marketing model name. That is not sufficient here: More optimization on training data can worsen generalization.

C:

A model can memorize training examples without generalizing. That is not sufficient here: More optimization on training data can worsen generalization.

Question 221

An AI operations team is testing two distinct capabilities. Failure (1): The team optimizes only the rate of blocked harmful prompts. Failure (2): Parsing drops headings and reading order from tagged PDFs. Which TWO specific corrections should the team apply?

  1. Assess both unsafe misses and harmful overblocking
  2. Use layout-preserving extraction and test alternatives
  3. Rebalance safety tests and adjust erroneous dialect associations
  4. Remove protected payloads from logging and enforce redaction tests
  5. Add an accessible name and keyboard activation to the control

Explanation

Correct Answers

A:

Correct for failure (1). Safety controls should not overblock permitted assistance. The proposed correction directly addresses that reported defect.

B:

Correct for failure (2). Inclusion requires robust input handling beyond a single document template. It independently corrects the second defect.

Incorrect Answers

C:

Fairness requires discriminating actual harm from linguistic variety. It does not complete the two decisions required in this question.

D:

Troubleshooting does not justify reproducing sensitive application records. It does not complete the two decisions required in this question.

E:

Inclusive design permits nonvisual navigation of the same task path. The stated pair of failures instead requires “Assess both unsafe misses and harmful overblocking” and “Use layout-preserving extraction and test alternatives”; this option cannot substitute for both.

Question 222

Troubleshooting report: The agent bypasses approval based on a purported urgent executive message. Which correction is MOST directly targeted to this failure?

  1. Audit and correct all cross-region ancillary data flows
  2. Enforce backend authorization and signed approval regardless of prompt wording
  3. Document feature lineage and remove unapproved decision features
  4. Permit calibrated caveats and qualified review paths

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. Accountable high-impact operations need verifiable authorized signatures. The targeted change is: Enforce backend authorization and signed approval regardless of prompt wording.

Incorrect Answers

A:

Privacy and compliance require controlling the full data path. In the stated case, troubleshooting report: The agent bypasses approval based on a purported urgent executive message; “Audit and correct all cross-region ancillary data flows” does not target that failure.

C:

Accountability requires interpretable reasons for consequential decisions. In the stated case, troubleshooting report: The agent bypasses approval based on a purported urgent executive message; “Document feature lineage and remove unapproved decision features” does not target that failure.

D:

Transparency reduces mistaken reliance on unsupported professional guarantees. In the stated case, troubleshooting report: The agent bypasses approval based on a purported urgent executive message; “Permit calibrated caveats and qualified review paths” does not target that failure.

Question 223

Troubleshooting report: The extractor returns an image tag called barcode without its value. Which correction is MOST directly targeted to this failure?

  1. Keep temporal and content provenance in the retrieval result
  2. Use barcode decoding and validate content
  3. Preserve actual section/page references through retrieval
  4. Configure supported language handling and test each script

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. A barcode is a structured visual symbol not ordinary OCR text. The targeted change is: Use barcode decoding and validate content.

Incorrect Answers

A:

Multimodal grounding should preserve how the answer maps to source media. In the stated case, troubleshooting report: The extractor returns an image tag called barcode without its value; “Keep temporal and content provenance in the retrieval result” does not target that failure.

C:

Structured layout and provenance let users verify a quoted clause. In the stated case, troubleshooting report: The extractor returns an image tag called barcode without its value; “Preserve actual section/page references through retrieval” does not target that failure.

D:

Recognition accuracy depends on script and language support. In the stated case, troubleshooting report: The extractor returns an image tag called barcode without its value; “Configure supported language handling and test each script” does not target that failure.

Question 224

An acceptance tester needs to validate: Disclose the automated nature and supported escalation route in the interface. Which observed result proves the intended behavior?

  1. Each policy answer citation resolves to the correct approved paragraph
  2. A first-use test displays clear AI identification and human contact option
  3. Dialect-balanced tests show benign requests are allowed without unsafe leakage
  4. An auditor can map each used input to an approved policy criterion

Explanation

Correct Answer

B:

Transparency helps users assess the scope and limitations of the assistance. The expected acceptance evidence is: A first-use test displays clear AI identification and human contact option. This observation directly tests the stated behavior.

Incorrect Answers

A:

Transparency lets users inspect evidence instead of accepting fluent text. This observation does not establish the required result: A first-use test displays clear AI identification and human contact option.

C:

Fairness requires discriminating actual harm from linguistic variety. This observation does not establish the required result: A first-use test displays clear AI identification and human contact option.

D:

Accountability requires interpretable reasons for consequential decisions. This observation does not establish the required result: A first-use test displays clear AI identification and human contact option.

Question 225

The approved design action is: Ground interpretation in visible labels and qualify uncertain numerical readings. Which test result would BEST establish that it worked?

  1. The summary preserves the order of observed demonstration steps and identifies gaps or uncertain actions
  2. The response distinguishes readable values from estimates
  3. A labeled test shows the chosen threshold achieves target precision/recall
  4. The output is a validated outdoor probability or category

Explanation

Correct Answer

B:

Visual reasoning is not proof of an exact value when axes are unclear. The expected acceptance evidence is: The response distinguishes readable values from estimates. This observation directly tests the stated behavior.

Incorrect Answers

A:

A single still cannot convey action order; sampled video frames may also omit fleeting intermediate steps. This observation does not establish the required result: The response distinguishes readable values from estimates.

C:

Confidence scores require calibration and contextual acceptance policy. This observation does not establish the required result: The response distinguishes readable values from estimates.

D:

Image-level classification is distinct from object localization. This observation does not establish the required result: The response distinguishes readable values from estimates.

Question 226

The approved design action is: Use key phrase extraction to identify salient text spans. Which test result would BEST establish that it worked?

  1. A consented test delivers equivalent meaning in the listener's language
  2. Test feedback returns relevant phrases found in the message
  3. Line-item totals and row relationships reconcile to the document
  4. The result contains product boxes in their correct shelf positions

Explanation

Correct Answer

B:

Key phrase extraction finds prominent topics without inventing new source material. The expected acceptance evidence is: Test feedback returns relevant phrases found in the message. This observation directly tests the stated behavior.

Incorrect Answers

A:

Speech translation spans audio input, semantic transfer and optional output audio. This observation does not establish the required result: Test feedback returns relevant phrases found in the message.

C:

Table extraction associates quantities, descriptions and prices by row. This observation does not establish the required result: Test feedback returns relevant phrases found in the message.

D:

Object detectors return classes and spatial boundaries. This observation does not establish the required result: Test feedback returns relevant phrases found in the message.

Question 227

Troubleshooting report: The app sends all historic messages on each trivial request. Which correction is MOST directly targeted to this failure?

  1. Review label provenance and downstream decision disparities
  2. Reduce sampling variability and validate outputs against a schema
  3. Apply context selection and retain necessary evidence only
  4. Split data by appropriate entities and evaluate on unseen cases

Explanation

Correct Answer

C:

This correction directly addresses the reported failure. Processing extra tokens increases cost and can affect response time. The targeted change is: Apply context selection and retain necessary evidence only.

Incorrect Answers

A:

Bias can enter through the targets themselves, not just inputs. In the stated case, troubleshooting report: The app sends all historic messages on each trivial request; “Review label provenance and downstream decision disparities” does not target that failure.

B:

Lower randomness often improves consistency but does not guarantee correctness. In the stated case, troubleshooting report: The app sends all historic messages on each trivial request; “Reduce sampling variability and validate outputs against a schema” does not target that failure.

D:

A model can memorize training examples without generalizing. In the stated case, troubleshooting report: The app sends all historic messages on each trivial request; “Split data by appropriate entities and evaluate on unseen cases” does not target that failure.

Question 228

A project needs the following AI behavior: A vision model reports objects with different confidence values. What should the designer implement?

  1. Ground interpretation in visible labels and qualify uncertain numerical readings
  2. Use evaluated confidence thresholds aligned to the business error cost
  3. Confirm the request includes actual supported image content and an image-capable deployment
  4. Choose detection only if spatial object locations are required

Explanation

Correct Answer

B:

Confidence scores require calibration and contextual acceptance policy. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Visual reasoning is not proof of an exact value when axes are unclear. That is not sufficient here: Confidence scores require calibration and contextual acceptance policy.

C:

A textual filename cannot convey pixels to a model. That is not sufficient here: Confidence scores require calibration and contextual acceptance policy.

D:

Classification predicts image-level categories while detection provides positions. That is not sufficient here: Confidence scores require calibration and contextual acceptance policy.

Question 229

An Azure AI Fundamentals solution has this requirement: A customer support platform needs to recommend a next troubleshooting article from prior successful cases. Which implementation most directly addresses it?

  1. Use a visual classification or detection model appropriate to whether location matters
  2. Use a relevance-ranking or recommendation model evaluated on resolved-case history
  3. Use named entity recognition with entity types
  4. Use invoice-oriented structured extraction with table row preservation

Explanation

Correct Answer

B:

Recommendations rank candidate items rather than simply generating arbitrary titles. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

The task chooses a label, with localization needed only when multiple signs appear. That is not sufficient here: Recommendations rank candidate items rather than simply generating arbitrary titles.

C:

NER locates and categorizes named entities in text. That is not sufficient here: Recommendations rank candidate items rather than simply generating arbitrary titles.

D:

Table extraction associates quantities, descriptions and prices by row. That is not sufficient here: Recommendations rank candidate items rather than simply generating arbitrary titles.

Question 230

A solution team has selected this control: Enforce the monetary threshold in the backend and require human authorization above it. Which observation is the most direct acceptance evidence?

  1. Above-limit test refunds remain pending until an authorized approver acts
  2. A cross-user test cannot retrieve another person's conversation fragments
  3. A compliance record links each exception to an owner and disposition
  4. A simulated timeout yields a failure status without a fabricated numeric balance

Explanation

Correct Answer

A:

A prompt-level request cannot override a transaction policy. The expected acceptance evidence is: Above-limit test refunds remain pending until an authorized approver acts. This observation directly tests the stated behavior.

Incorrect Answers

B:

Privacy includes preventing cross-session memory disclosure. This observation does not establish the required result: Above-limit test refunds remain pending until an authorized approver acts.

C:

Self-attestation is not independent proof of regulatory controls. This observation does not establish the required result: Above-limit test refunds remain pending until an authorized approver acts.

D:

A fluent generation cannot substitute for missing authoritative transaction data. This observation does not establish the required result: Above-limit test refunds remain pending until an authorized approver acts.

Microsoft Azure AI Fundamentals (AI-901) Practice Tests