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 30

Question 290

An Azure AI Fundamentals solution has this requirement: An application needs to read a printed serial number from a device photograph. Which implementation most directly addresses it?

  1. Choose detection only if spatial object locations are required
  2. Use optical character recognition for image text
  3. Use region-aware captioning or a cropped image focused on that location
  4. Apply bounded retries and respect the service's throughput limit

Explanation

Correct Answer

B:

OCR extracts characters visible in pixels. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Classification predicts image-level categories while detection provides positions. That is not sufficient here: OCR extracts characters visible in pixels.

C:

Specific regions can carry details missed by a whole-image caption. That is not sufficient here: OCR extracts characters visible in pixels.

D:

Too many concurrent requests can exceed resource quota. That is not sufficient here: OCR extracts characters visible in pixels.

Question 291

An engineering review documents this requirement: A visual data extraction workflow must record where in an image a value came from. Select the most appropriate solution.

  1. Use text-to-image generation and keep the input photo out of the edit request
  2. Request an image caption or supported multimodal description
  3. Provide the actual image to a capable multimodal or OCR pipeline
  4. Preserve region coordinates with extracted text or field evidence

Explanation

Correct Answer

D:

Spatial provenance supports later verification of recognition results. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Generation from text is distinct from image editing. That is not sufficient here: Spatial provenance supports later verification of recognition results.

B:

Captioning produces a sentence-level description rather than a coordinate list. That is not sufficient here: Spatial provenance supports later verification of recognition results.

C:

Visual evidence is needed to read sign content. That is not sufficient here: Spatial provenance supports later verification of recognition results.

Question 292

Troubleshooting report: The team lowers the microphone sample rate to change speaking speed. Which correction is MOST directly targeted to this failure?

  1. Select an appropriate audio-input model and upload the audio correctly
  2. Supply an authorized identity and matching resource configuration
  3. Control synthesis prosody in SSML
  4. Adjust recognition vocabulary and test the target terms

Explanation

Correct Answer

C:

This correction directly addresses the reported failure. Voice selection and prosody can influence delivery without changing the text. The targeted change is: Control synthesis prosody in SSML.

Incorrect Answers

A:

Model capabilities and accepted payload types differ across deployments. In the stated case, troubleshooting report: The team lowers the microphone sample rate to change speaking speed; “Select an appropriate audio-input model and upload the audio correctly” does not target that failure.

B:

Audio capture success does not establish remote service authentication. In the stated case, troubleshooting report: The team lowers the microphone sample rate to change speaking speed; “Supply an authorized identity and matching resource configuration” does not target that failure.

D:

Domain-specific words may require biasing or adaptation beyond a generic recognizer. In the stated case, troubleshooting report: The team lowers the microphone sample rate to change speaking speed; “Adjust recognition vocabulary and test the target terms” does not target that failure.

Question 293

The approved design action is: Treat user content as untrusted and maintain privileged instruction boundaries. Which test result would BEST establish that it worked?

  1. The target model deploys in the approved region and meets residency controls
  2. Conflict tests keep disallowed actions blocked
  3. Injection attempts fail to access protected secrets or tools
  4. The same app points to the intended test or production resource by configuration

Explanation

Correct Answer

C:

A user cannot use text to elevate permissions beyond the application policy. The expected acceptance evidence is: Injection attempts fail to access protected secrets or tools. This observation directly tests the stated behavior.

Incorrect Answers

A:

Not every model or feature is available in every location. This observation does not establish the required result: Injection attempts fail to access protected secrets or tools.

B:

Developer and system constraints govern allowed operations over user requests. This observation does not establish the required result: Injection attempts fail to access protected secrets or tools.

D:

A correct code path can still target the wrong resource after migration. This observation does not establish the required result: Injection attempts fail to access protected secrets or tools.

Question 294

Troubleshooting report: The team ignores validation and trains indefinitely. Which correction is MOST directly targeted to this failure?

  1. Use early stopping and investigate excessive model capacity
  2. Use a speech synthesis endpoint and compatible output format
  3. Use a multimodal-capable deployment and supported message format
  4. Choose a recognized audio-input interface and accepted encoding

Explanation

Correct Answer

A:

This correction directly addresses the reported failure. More optimization on training data can worsen generalization. The targeted change is: Use early stopping and investigate excessive model capacity.

Incorrect Answers

B:

Speech synthesis generates audio from text rather than transcribing audio. In the stated case, troubleshooting report: The team ignores validation and trains indefinitely; “Use a speech synthesis endpoint and compatible output format” does not target that failure.

C:

Multimodal models can reason over more than one supported input modality. In the stated case, troubleshooting report: The team ignores validation and trains indefinitely; “Use a multimodal-capable deployment and supported message format” does not target that failure.

D:

Model capabilities must include the required audio modality before the client can obtain speech-to-text results. In the stated case, troubleshooting report: The team ignores validation and trains indefinitely; “Choose a recognized audio-input interface and accepted encoding” does not target that failure.

Question 295

An AI operations team is testing two distinct capabilities. Failure (1): The only test file is a noise-free short sample. Failure (2): All text is attributed to the call-center agent. Which TWO specific corrections should the team apply?

  1. Add realistic noise tests and a clarification fallback
  2. Enable supported diarization and validate turn assignments
  3. Use the correct audio encoding and recognized input format
  4. Apply a confirmation prompt and constrained recognition
  5. Gate side effects on final recognition and explicit confirmation

Explanation

Correct Answers

A:

Correct for failure (1). Signal quality directly affects recognizer error rates. The proposed correction directly addresses that reported defect.

B:

Correct for failure (2). Speaker diarization separates turns and identities at a conversation level. It independently corrects the second defect.

Incorrect Answers

C:

Speech recognition requires the service to decode audio correctly. The stated pair of failures instead requires “Add realistic noise tests and a clarification fallback” and “Enable supported diarization and validate turn assignments”; this option cannot substitute for both.

D:

Speech recognition uncertainty is higher for confusable numeric strings. The stated pair of failures instead requires “Add realistic noise tests and a clarification fallback” and “Enable supported diarization and validate turn assignments”; this option cannot substitute for both.

E:

Interim hypotheses can change as more audio arrives. It does not complete the two decisions required in this question.

Question 296

An Azure AI Fundamentals solution has this requirement: A team asks whether lower temperature makes a model factually correct. Which implementation most directly addresses it?

  1. Run adversarial model evaluations and verify that configured content-safety guardrails enforce the stated high-severity block criteria
  2. Distinguish generation variability controls from evidence-based factual accuracy
  3. Require explicit confirmation and backend authorization before tool execution
  4. Limit context to necessary messages and apply safe summaries or retrieval

Explanation

Correct Answer

B:

Sampling parameters influence diversity but do not verify knowledge. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Adversarial testing must measure actual configured policy enforcement rather than treating a prompt instruction as a guaranteed safety boundary. That is not sufficient here: Sampling parameters influence diversity but do not verify knowledge.

C:

High-impact side effects need permissions independent of conversational language. That is not sufficient here: Sampling parameters influence diversity but do not verify knowledge.

D:

Long irrelevant context consumes tokens and may increase latency. That is not sufficient here: Sampling parameters influence diversity but do not verify knowledge.

Question 297

Troubleshooting report: The team evaluates mean squared distance of vehicle IDs. Which correction is MOST directly targeted to this failure?

  1. Add bounded conversation state and retrieved evidence
  2. Train a binary failure classifier using historical labeled cases
  3. Train or tune a contextual recommender and evaluate precision at rank
  4. Interpret the actual image first and synthesize the verified description

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. A failure probability or class is a discrete decision under a defined horizon. The targeted change is: Train a binary failure classifier using historical labeled cases.

Incorrect Answers

A:

The assistant should incorporate context while remaining connected to reliable sources. In the stated case, troubleshooting report: The team evaluates mean squared distance of vehicle IDs; “Add bounded conversation state and retrieved evidence” does not target that failure.

C:

Recommendations rank candidate items rather than simply generating arbitrary titles. In the stated case, troubleshooting report: The team evaluates mean squared distance of vehicle IDs; “Train or tune a contextual recommender and evaluate precision at rank” does not target that failure.

D:

Multimodal input can describe the graphic and speech output can read that description. In the stated case, troubleshooting report: The team evaluates mean squared distance of vehicle IDs; “Interpret the actual image first and synthesize the verified description” does not target that failure.

Question 298

Troubleshooting report: Raw identifiable frames are publicly attached to all summaries. Which correction is MOST directly targeted to this failure?

  1. Mask or exclude unauthorized identity details before publishing
  2. Use layout-preserving extraction and test alternatives
  3. Document feature lineage and remove unapproved decision features
  4. Run the frozen adversarial and ordinary acceptance suite against the candidate

Explanation

Correct Answer

A:

This correction directly addresses the reported failure. Privacy applies to image content as well as text transcripts. The targeted change is: Mask or exclude unauthorized identity details before publishing.

Incorrect Answers

B:

Inclusion requires robust input handling beyond a single document template. In the stated case, troubleshooting report: Raw identifiable frames are publicly attached to all summaries; “Use layout-preserving extraction and test alternatives” does not target that failure.

C:

Accountability requires interpretable reasons for consequential decisions. In the stated case, troubleshooting report: Raw identifiable frames are publicly attached to all summaries; “Document feature lineage and remove unapproved decision features” does not target that failure.

D:

Reliability includes preventing behavior regression across model versions. In the stated case, troubleshooting report: Raw identifiable frames are publicly attached to all summaries; “Run the frozen adversarial and ordinary acceptance suite against the candidate” does not target that failure.

Question 299

An AI operations team is testing two distinct capabilities. Failure (1): The invoice extractor is applied unchanged to every contract. Failure (2): The checkbox is discarded during plain text extraction. Which TWO specific corrections should the team apply?

  1. Introduce classification and type-specific routing
  2. Poll the documented operation status and handle failures
  3. Enable supported layout and field extraction features
  4. Keep transcript anchors and validate extracted tasks
  5. Include selection-mark processing in the analyzer

Explanation

Correct Answers

A:

Correct for failure (1). Different document kinds have distinct field schemas and review paths. The proposed correction directly addresses that reported defect.

E:

Correct for failure (2). Checkbox state is a structured signal rather than arbitrary prose. It independently corrects the second defect.

Incorrect Answers

B:

Long-running media analysis may be asynchronous. It does not complete the two decisions required in this question.

C:

Ordinary plain-text OCR may lose table relationships and checkbox states. The stated pair of failures instead requires “Introduce classification and type-specific routing” and “Include selection-mark processing in the analyzer”; this option cannot substitute for both.

D:

Transcription and summarization are separate results with different validation. The stated pair of failures instead requires “Introduce classification and type-specific routing” and “Include selection-mark processing in the analyzer”; this option cannot substitute for both.

Question 300

An Azure-based AI team must satisfy this objective: A web app exposes a model endpoint directly to untrusted browser clients. Which design choice is best?

  1. Run adversarial model evaluations and verify that configured content-safety guardrails enforce the stated high-severity block criteria
  2. Implement a network timeout/retry strategy that avoids duplicate tool effects
  3. Place the current request in the supported user message field
  4. Call the deployed model through an authenticated application backend

Explanation

Correct Answer

D:

Backend controls protect credentials, quotas and data policy. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Adversarial testing must measure actual configured policy enforcement rather than treating a prompt instruction as a guaranteed safety boundary. That is not sufficient here: Backend controls protect credentials, quotas and data policy.

B:

Client retries should distinguish read requests from side-effecting tool calls. That is not sufficient here: Backend controls protect credentials, quotas and data policy.

C:

The inference service needs task content to generate a relevant answer. That is not sufficient here: Backend controls protect credentials, quotas and data policy.

Microsoft Azure AI Fundamentals (AI-901) Practice Tests