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 22

Question 210

An Azure AI solution uses the following approved control: Call a supported text sentiment-analysis feature and interpret returned polarity. What observable result should pass QA?

  1. Labeled sample complaints and praise yield expected sentiment categories
  2. The acronym is spoken according to the approved pronunciation sample
  3. A labeled command yields readable sentence boundaries
  4. The output audio reads the approved sentence intelligibly

Explanation

Correct Answer

A:

Sentiment output classifies attitudes expressed in the submitted text. The expected acceptance evidence is: Labeled sample complaints and praise yield expected sentiment categories. This observation directly tests the stated behavior.

Incorrect Answers

B:

SSML can guide pronunciation beyond plain-text guessing. This observation does not establish the required result: Labeled sample complaints and praise yield expected sentiment categories.

C:

Speech text may require punctuation inference distinct from word recognition. This observation does not establish the required result: Labeled sample complaints and praise yield expected sentiment categories.

D:

Speech synthesis generates audio waveforms from supplied text. This observation does not establish the required result: Labeled sample complaints and praise yield expected sentiment categories.

Question 211

A project needs the following AI behavior: A custom analyzer produces values but no field confidence for an automation pipeline. What should the designer implement?

  1. Wait for the analysis operation to reach a terminal state before reading results
  2. Use a video-capable Content Understanding analyzer with timestamps and supported fields
  3. Use array-typed field extraction rather than one scalar field
  4. Enable supported document-field confidence and grounding, then calibrate field-specific review thresholds using labeled samples

Explanation

Correct Answer

D:

Content Understanding confidence applies to supported document analyzers; confidence estimates alone do not prove field correctness.

Incorrect Answers

A:

Long-running media analysis may be asynchronous. That is not sufficient here: Content Understanding confidence applies to supported document analyzers; confidence estimates alone do not prove field correctness.

B:

Multimodal extraction draws on visual and spoken evidence over time. That is not sufficient here: Content Understanding confidence applies to supported document analyzers; confidence estimates alone do not prove field correctness.

C:

Repeated records require structured lists with item-level associations. That is not sufficient here: Content Understanding confidence applies to supported document analyzers; confidence estimates alone do not prove field correctness.

Question 212

A solution team has selected this control: Choose a speech-recognition model or audio-capable multimodal endpoint with a documented audio-input contract. Which observation is the most direct acceptance evidence?

  1. A peak-load forecast matches the configured token/request limits or identifies an approved capacity increase
  2. The returned text includes the photographed serial number
  3. The model/deployment capability check and a spoken-audio test confirm that this endpoint supports transcription
  4. A user request cannot override the controlling system-level instructions in tests

Explanation

Correct Answer

C:

Model capabilities must include the required audio modality before the client can obtain speech-to-text results. The expected acceptance evidence is: The model/deployment capability check and a spoken-audio test confirm that this endpoint supports transcription. This observation directly tests the stated behavior.

Incorrect Answers

A:

Quota sizing is a deployment-capacity decision; retrying a predictable sustained excess cannot add model throughput. This observation does not establish the required result: The model/deployment capability check and a spoken-audio test confirm that this endpoint supports transcription.

B:

OCR produces readable text from pixels. This observation does not establish the required result: The model/deployment capability check and a spoken-audio test confirm that this endpoint supports transcription.

D:

Instruction roles help distinguish application policy from user task text. This observation does not establish the required result: The model/deployment capability check and a spoken-audio test confirm that this endpoint supports transcription.

Question 213

An Azure-based AI team must satisfy this objective: A speech app is using text-to-speech but no sound plays on an embedded device. Which design choice is best?

  1. Check playback device selection and the synthesized audio format
  2. Use speech-to-text recognition on a supported audio input stream
  3. Escape XML special characters and validate SSML syntax
  4. Match the speech resource region or endpoint to the SDK configuration

Explanation

Correct Answer

A:

Successful synthesis does not guarantee local audio is routed to a speaker. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

B:

Speech recognition converts waveform features to text tokens. That is not sufficient here: Successful synthesis does not guarantee local audio is routed to a speaker.

C:

SSML is XML and malformed markup breaks the synthesis request. That is not sufficient here: Successful synthesis does not guarantee local audio is routed to a speaker.

D:

Region or endpoint mismatch prevents requests reaching the correct resource. That is not sufficient here: Successful synthesis does not guarantee local audio is routed to a speaker.

Question 214

Troubleshooting report: The application overwrites the table amount with a guessed newer value. Which correction is MOST directly targeted to this failure?

  1. Preserve conflict flags and apply approved precedence rules
  2. Keep transcript anchors and validate extracted tasks
  3. Mark the field unreadable and request manual correction
  4. Keep temporal and content provenance in the retrieval result

Explanation

Correct Answer

A:

This correction directly addresses the reported failure. A model should not silently choose among competing contractual evidence. The targeted change is: Preserve conflict flags and apply approved precedence rules.

Incorrect Answers

B:

Transcription and summarization are separate results with different validation. In the stated case, troubleshooting report: The application overwrites the table amount with a guessed newer value; “Keep transcript anchors and validate extracted tasks” does not target that failure.

C:

Low-quality inputs can make extraction genuinely uncertain. In the stated case, troubleshooting report: The application overwrites the table amount with a guessed newer value; “Mark the field unreadable and request manual correction” does not target that failure.

D:

Multimodal grounding should preserve how the answer maps to source media. In the stated case, troubleshooting report: The application overwrites the table amount with a guessed newer value; “Keep temporal and content provenance in the retrieval result” does not target that failure.

Question 215

Troubleshooting report: The team approves deployment based only on a few successful demos. Which correction is MOST directly targeted to this failure?

  1. Add an accessible name and keyboard activation to the control
  2. Use layout-preserving extraction and test alternatives
  3. Run the frozen adversarial and ordinary acceptance suite against the candidate
  4. Scope the token and backend authorization to necessary actions

Explanation

Correct Answer

C:

This correction directly addresses the reported failure. Reliability includes preventing behavior regression across model versions. The targeted change is: Run the frozen adversarial and ordinary acceptance suite against the candidate.

Incorrect Answers

A:

Inclusive design permits nonvisual navigation of the same task path. In the stated case, troubleshooting report: The team approves deployment based only on a few successful demos; “Add an accessible name and keyboard activation to the control” does not target that failure.

B:

Inclusion requires robust input handling beyond a single document template. In the stated case, troubleshooting report: The team approves deployment based only on a few successful demos; “Use layout-preserving extraction and test alternatives” does not target that failure.

D:

Least privilege reduces damage from compromised prompts or tool output. In the stated case, troubleshooting report: The team approves deployment based only on a few successful demos; “Scope the token and backend authorization to necessary actions” does not target that failure.

Question 216

Troubleshooting report: Every failed image call is retried immediately with no backoff. Which correction is MOST directly targeted to this failure?

  1. Apply orientation metadata or rotate pixels consistently
  2. Map the task to a whole-image classifier
  3. Add confidence gating and supporting evidence
  4. Limit concurrency and honor retry guidance

Explanation

Correct Answer

D:

This correction directly addresses the reported failure. Too many concurrent requests can exceed resource quota. The targeted change is: Limit concurrency and honor retry guidance.

Incorrect Answers

A:

Rotated inputs can degrade detection or coordinate interpretation. In the stated case, troubleshooting report: Every failed image call is retried immediately with no backoff; “Apply orientation metadata or rotate pixels consistently” does not target that failure.

B:

Image-level classification is distinct from object localization. In the stated case, troubleshooting report: Every failed image call is retried immediately with no backoff; “Map the task to a whole-image classifier” does not target that failure.

C:

Generated visual interpretations can be wrong even when phrased fluently. In the stated case, troubleshooting report: Every failed image call is retried immediately with no backoff; “Add confidence gating and supporting evidence” does not target that failure.

Question 217

A solution team has selected this control: Use grounded medical content, high-risk detection and appropriate expert escalation. Which observation is the most direct acceptance evidence?

  1. A cross-user test cannot retrieve another person's conversation fragments
  2. Out-of-scope questions trigger qualified language and reference suggestions
  3. An unsafe self-treatment case triggers a safe refusal or professional referral
  4. Accent-specific word error rates meet stated thresholds or trigger alternatives

Explanation

Correct Answer

C:

Safety mechanisms should reduce unsupported high-impact recommendations. The expected acceptance evidence is: An unsafe self-treatment case triggers a safe refusal or professional referral. This observation directly tests the stated behavior.

Incorrect Answers

A:

Privacy includes preventing cross-session memory disclosure. This observation does not establish the required result: An unsafe self-treatment case triggers a safe refusal or professional referral.

B:

Honesty requires acknowledging when a claim cannot be substantiated. This observation does not establish the required result: An unsafe self-treatment case triggers a safe refusal or professional referral.

D:

Inclusive voice systems require performance evaluation beyond a single accent. This observation does not establish the required result: An unsafe self-treatment case triggers a safe refusal or professional referral.

Question 218

Troubleshooting report: An exception handler substitutes a model-estimated amount for an unavailable balance. Which correction is MOST directly targeted to this failure?

  1. Scope the token and backend authorization to necessary actions
  2. Run the frozen adversarial and ordinary acceptance suite against the candidate
  3. Propagate typed tool errors and use an explicit unavailable response
  4. Enforce backend authorization and signed approval regardless of prompt wording

Explanation

Correct Answer

C:

This correction directly addresses the reported failure. A fluent generation cannot substitute for missing authoritative transaction data. The targeted change is: Propagate typed tool errors and use an explicit unavailable response.

Incorrect Answers

A:

Least privilege reduces damage from compromised prompts or tool output. In the stated case, troubleshooting report: An exception handler substitutes a model-estimated amount for an unavailable balance; “Scope the token and backend authorization to necessary actions” does not target that failure.

B:

Reliability includes preventing behavior regression across model versions. In the stated case, troubleshooting report: An exception handler substitutes a model-estimated amount for an unavailable balance; “Run the frozen adversarial and ordinary acceptance suite against the candidate” does not target that failure.

D:

Accountable high-impact operations need verifiable authorized signatures. In the stated case, troubleshooting report: An exception handler substitutes a model-estimated amount for an unavailable balance; “Enforce backend authorization and signed approval regardless of prompt wording” does not target that failure.

Question 219

An AI operations team is testing two distinct capabilities. Failure (1): The service spells the acronym unpredictably in generic synthesis. Failure (2): The application permanently archives every waveform for debugging. Which TWO specific corrections should the team apply?

  1. Use the correct audio encoding and recognized input format
  2. Configure retention and delete raw recordings when no longer necessary
  3. Add an authorized synthesis stage for the answer
  4. Add the suitable pronunciation markup or lexicon
  5. Choose the correct TTS voice from the supported catalog

Explanation

Correct Answers

B:

Correct for failure (2). Speech workloads still process personal data and require controlled lifecycle. It independently corrects the second defect.

D:

Correct for failure (1). SSML can guide pronunciation beyond plain-text guessing. The proposed correction directly addresses that reported defect.

Incorrect Answers

A:

Speech recognition requires the service to decode audio correctly. The stated pair of failures instead requires “Configure retention and delete raw recordings when no longer necessary” and “Add the suitable pronunciation markup or lexicon”; this option cannot substitute for both.

C:

Input transcription and output synthesis are distinct steps. It does not complete the two decisions required in this question.

E:

Voice timbre is a synthesis configuration, not speech recognition output. The stated pair of failures instead requires “Configure retention and delete raw recordings when no longer necessary” and “Add the suitable pronunciation markup or lexicon”; this option cannot substitute for both.

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.

Microsoft Azure AI Fundamentals (AI-901) Practice Tests