AWS Certified AI Practitioner AIF-C01 Practice Test 3
Question 20
A demand model performs nearly perfectly on training records but poorly on a new seasonal holdout set. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- A reviewed design record approves this competing approach: Skip validation because the model finished deploying successfully
- Holdout error converges with training performance on representative future-period samples
- A component test invokes this alternate approach successfully: Increase the number of training epochs without checking holdout loss
- A system inventory lists this competing approach as active: Report training-set accuracy alone as proof of future accuracy
Explanation
Correct Answer
B:Holdout error converges with training performance on representative future-period samples. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Deployment completion is independent of predictive generalization Design approval is not evidence that the alternative meets the actual requirement.
C:More optimization can increase overfitting when generalization is already weak A component invocation does not validate the necessary end-to-end outcome.
D:Training fit does not measure performance on unseen seasonal samples An active setting cannot independently prove the requested business outcome.
Question 21
A manufacturer wants to classify defective and acceptable products using photographs of the assembly line. During a design review, someone proposes Use text sentiment analysis of machine serial numbers. Which technical objection is specifically relevant to this proposal?
- The decisive signal is visual image content rather than plain-text or speech
- Transcription converts audio speech and cannot classify pixels
- Sentiment on identifiers does not inspect visual product defects
- Audio output does not identify the defect in the incoming image
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Use text sentiment analysis of machine serial numbers. Sentiment on identifiers does not inspect visual product defects
Incorrect Answers
A:The decisive signal is visual image content rather than plain-text or speech This explains the recommended design, but it is not the requested diagnosis of Use text sentiment analysis of machine serial numbers.
B:This limitation concerns Use speech-to-text transcription to interpret conveyor photographs; it does not diagnose the proposed Use text sentiment analysis of machine serial numbers approach.
D:This limitation concerns Use speech synthesis to report labels without classifying images; it does not diagnose the proposed Use text sentiment analysis of machine serial numbers approach.
Question 22
An online checkout system must return a fraud score while the customer is still waiting. During a design review, someone proposes Require humans to retrain the model before every purchase. Which technical objection is specifically relevant to this proposal?
- An interactive decision requires an endpoint meeting the application's latency budget
- Model training is separate from serving each inference request
- Translating purchase descriptions is not a fraud score
- Nightly batch results arrive after the checkout decision window
Explanation
Correct Answer
B:This is the actual failure mode of the proposed approach: Require humans to retrain the model before every purchase. Model training is separate from serving each inference request
Incorrect Answers
A:An interactive decision requires an endpoint meeting the application's latency budget This explains the recommended design, but it is not the requested diagnosis of Require humans to retrain the model before every purchase.
C:This limitation concerns Use a translation pipeline as the fraud scoring endpoint; it does not diagnose the proposed Require humans to retrain the model before every purchase approach.
D:This limitation concerns Schedule a single nightly batch job for all checkout requests; it does not diagnose the proposed Require humans to retrain the model before every purchase approach.
Question 23
A model training project fails because currency fields contain strings and missing values are inconsistent. Which option BEST meets the business requirement?
- Tune output temperature on an existing generative model to repair a broken CSV
- Assume deployment monitoring replaces initial data preparation
- Skip training-data validation and rely only on more expensive compute
- Clean and validate training data before fitting the model
Explanation
Correct Answer
D:Preprocessing resolves malformed values and missing-data treatment that otherwise corrupt features The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Sampling temperature does not repair tabular training data quality This prevents the approach from meeting the stated requirement.
B:Monitoring after deployment does not clean malformed training records This prevents the approach from meeting the stated requirement.
C:Extra compute cannot guarantee consistent features or valid numeric inputs This prevents the approach from meeting the stated requirement.
Question 24
A manufacturer wants to classify defective and acceptable products using photographs of the assembly line. Which option BEST meets the business requirement?
- Use speech synthesis to report labels without classifying images
- Use text sentiment analysis of machine serial numbers
- Use computer-vision image classification trained or adapted for visual defects
- Use speech-to-text transcription to interpret conveyor photographs
Explanation
Correct Answer
C:The decisive signal is visual image content rather than plain-text or speech The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Audio output does not identify the defect in the incoming image This prevents the approach from meeting the stated requirement.
B:Sentiment on identifiers does not inspect visual product defects This prevents the approach from meeting the stated requirement.
D:Transcription converts audio speech and cannot classify pixels This prevents the approach from meeting the stated requirement.
Question 25
An operations team must process a million stored images overnight rather than respond to user uploads instantly. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- A component test invokes this alternate approach successfully: Train a new model for each uploaded image
- A reviewed design record approves this competing approach: Replace image inference with an audio conversion service
- The full batch completes by the agreed processing deadline with recorded output coverage
- A system inventory lists this competing approach as active: Choose only synchronous low-latency requests for each image
Explanation
Correct Answer
C:The full batch completes by the agreed processing deadline with recorded output coverage. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Re-training per input is not normal inference and adds avoidable cost A component invocation does not validate the necessary end-to-end outcome.
B:Audio conversion does not produce requested visual predictions Design approval is not evidence that the alternative meets the actual requirement.
D:Interactive calls add unnecessary per-request pressure when offline processing is allowed An active setting cannot independently prove the requested business outcome.
Question 26
A simulated warehouse robot learns which route to choose from delayed rewards for delivery time and collisions. During a design review, someone proposes Use standard regression to output a single shipping price. Which technical objection is specifically relevant to this proposal?
- Recognizing labels is not a learned navigation policy driven by rewards
- Clustering does not optimize sequences of robot actions against rewards
- A price estimate does not determine route actions over time
- The agent learns actions from rewards and environmental feedback across episodes
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Use standard regression to output a single shipping price. A price estimate does not determine route actions over time
Incorrect Answers
A:This limitation concerns Use supervised OCR to read every shelf label as the entire control policy; it does not diagnose the proposed Use standard regression to output a single shipping price approach.
B:This limitation concerns Use unsupervised clustering to assign static product categories; it does not diagnose the proposed Use standard regression to output a single shipping price approach.
D:The agent learns actions from rewards and environmental feedback across episodes This explains the recommended design, but it is not the requested diagnosis of Use standard regression to output a single shipping price.
Question 27
An email service has thousands of messages labeled spam or legitimate and needs to classify future messages. A pilot uses this approach instead: Use unsupervised clustering without consulting existing labels. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Regression predicts numerical quantities rather than the specified discrete spam class
- Labels identify the target classes and allow the classifier to learn a decision boundary
- Clustering groups by similarity without directly learning the recorded spam labels
- A reward-driven control policy does not directly exploit the provided labeled email classes
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Use unsupervised clustering without consulting existing labels. Clustering groups by similarity without directly learning the recorded spam labels
Incorrect Answers
A:This limitation concerns Use regression to predict a continuous dollar amount; it does not diagnose the proposed Use unsupervised clustering without consulting existing labels approach.
B:Labels identify the target classes and allow the classifier to learn a decision boundary This explains the recommended design, but it is not the requested diagnosis of Use unsupervised clustering without consulting existing labels.
D:This limitation concerns Use reinforcement learning with rewards for warehouse robots; it does not diagnose the proposed Use unsupervised clustering without consulting existing labels approach.
Question 28
An online checkout system must return a fraud score while the customer is still waiting. A pilot uses this approach instead: Schedule a single nightly batch job for all checkout requests. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Translating purchase descriptions is not a fraud score
- An interactive decision requires an endpoint meeting the application's latency budget
- Model training is separate from serving each inference request
- Nightly batch results arrive after the checkout decision window
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Schedule a single nightly batch job for all checkout requests. Nightly batch results arrive after the checkout decision window
Incorrect Answers
A:This limitation concerns Use a translation pipeline as the fraud scoring endpoint; it does not diagnose the proposed Schedule a single nightly batch job for all checkout requests approach.
B:An interactive decision requires an endpoint meeting the application's latency budget This explains the recommended design, but it is not the requested diagnosis of Schedule a single nightly batch job for all checkout requests.
C:This limitation concerns Require humans to retrain the model before every purchase; it does not diagnose the proposed Schedule a single nightly batch job for all checkout requests approach.
Question 29
A hotel wants a conversational bot that collects check-in date and room type through voice or text turns. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- The bot captures the required slots and correctly invokes an approved reservation workflow
- A component test invokes this alternate approach successfully: Use Amazon Translate alone to track hotel reservation state
- A system inventory lists this competing approach as active: Use Amazon Polly alone as the complete booking dialog engine
- A reviewed design record approves this competing approach: Use Amazon Rekognition to infer date and room type from speech
Explanation
Correct Answer
A:The bot captures the required slots and correctly invokes an approved reservation workflow. This checks the business result rather than just a configured feature.
Incorrect Answers
B:Translation does not orchestrate multi-turn reservation fields A component invocation does not validate the necessary end-to-end outcome.
C:Polly speaks text but does not manage user intents or slot collection An active setting cannot independently prove the requested business outcome.
D:Image analysis is not a spoken reservation dialog manager Design approval is not evidence that the alternative meets the actual requirement.
Question 30
A data analyst receives transaction tables, customer chat text and store-camera photos for one research initiative. Which option BEST meets the business requirement?
- Treat rows as structured tabular data and text/images as unstructured modalities
- Assume labeled and unlabeled mean the same thing as structured and unstructured
- Treat every numeric transaction column as a video frame
- Classify every input as tabular merely because it is stored in S3
Explanation
Correct Answer
A:Each modality has different representation and preprocessing requirements The specified scenario makes this the most appropriate approach.
Incorrect Answers
B:Target-label availability and data structure are separate properties This prevents the approach from meeting the stated requirement.
C:Numeric table columns are not image or video pixels This prevents the approach from meeting the stated requirement.
D:Object-storage location does not make image pixels or natural text tabular This prevents the approach from meeting the stated requirement.
