AWS MLA-C01 Practice Question Bank
334 unique multiple-choice questions for the AWS Certified Machine Learning Engineer – Associate (MLA-C01) exam—with detailed explanations for every answer and distractor.
In short
Unofficial practice aligned to domain weights (28 / 26 / 22 / 24). Expand each question to reveal the answer and rationale. Verify against the latest official exam guide.
Exam at a glance
| Scored questions | 50 |
|---|---|
| Total (incl. unscored) | 65 |
| Time | 130 minutes |
| Passing score | 720 / 1000 (scaled) |
| Domain | Weight |
|---|---|
| 1 — Data Preparation | 28% |
| 2 — ML Model Development | 26% |
| 3 — Deployment & Orchestration | 22% |
| 4 — Monitoring, Maintenance & Security | 24% |
AWS markets this as Machine Learning Engineer – Associate (MLA-C01), not a separate “MLOps Associate” badge. It covers building, deploying, orchestrating, monitoring, and securing ML workloads on SageMaker and related services.
Jump to a domain
- Domain 1: Data Preparation for ML (Q1–93)
- Domain 2: ML Model Development (Q94–180)
- Domain 3: Deployment and Orchestration (Q181–253)
- Domain 4: Monitoring, Maintenance, and Security (Q254–334)
Questions
Click a question to expand the answer and explanations.
Domain 1: Data Preparation for ML
Question 1. A data team stores 2 TB of clickstream logs accessed mostly by column (user_id, event_type, timestamp). They run nightly Spark jobs on EMR and occasional ad-hoc Athena queries. Which storage format best balances cost and query performance?
- A) JSON files in Amazon S3 with gzip compression
- B) Apache Parquet with Snappy compression partitioned by date
- C) CSV without compression for human readability
- D) Amazon EBS gp3 volumes attached to a single EC2 instance
Correct answer: B
Why this is correct: Parquet is columnar: Spark and Athena read only needed columns, reducing I/O versus row formats like JSON/CSV. Snappy gives fast compression/decompression suitable for big data. Partitioning by date limits scan scope. EBS on one instance does not scale for shared analytics lakes.
Why the distractors are incorrect:
- A) JSON files in Amazon S3 with gzip compression — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) CSV without compression for human readability — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Amazon EBS gp3 volumes attached to a single EC2 instance — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 2. An ML engineer must land real-time fraud features into a feature store while also archiving raw events for compliance. Which architecture satisfies both paths?
- A) Kinesis Data Streams → Lambda → SageMaker Feature Store; parallel Firehose delivery to S3
- B) S3 PUT events only, with no streaming layer
- C) Amazon RDS Multi-AZ as the sole ingestion point
- D) AWS Snowball Edge for every transaction
Correct answer: A
Why this is correct: Kinesis ingests streams; Lambda writes online features; Firehose archives to S3 for audit/replay.
Why the distractors are incorrect:
- B) S3 PUT events only, with no streaming layer — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) Amazon RDS Multi-AZ as the sole ingestion point — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) AWS Snowball Edge for every transaction — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 3. Before training a credit model, you must detect label imbalance across protected groups. Which AWS tool is purpose-built for pre-training bias metrics (e.g., CI, DPL)?
- A) Amazon Macie
- B) SageMaker Clarify pre-training bias analysis
- C) AWS WAF
- D) Amazon CloudFront
Correct answer: B
Why this is correct: Clarify computes CI, DPL, and related pre-training metrics. Macie finds sensitive data; WAF/CloudFront are edge security/CDN.
Why the distractors are incorrect:
- A) Amazon Macie — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) AWS WAF — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Amazon CloudFront — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 4. A tabular regression problem has 50k rows and strong linear relationships. Fast baseline on SageMaker?
- A) Linear Learner
- B) Object Detection
- C) BlazingText Word2Vec only
- D) Seq2Seq
Correct answer: A
Why this is correct: Linear Learner fits tabular regression/classification. Others target vision, embeddings, or sequences.
Why the distractors are incorrect:
- B) Object Detection — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) BlazingText Word2Vec only — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Seq2Seq — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 5. Hyperparameter search across 20 continuous parameters with limited budget—smarter than naive grid?
- A) SageMaker Automatic Model Tuning (Bayesian optimization)
- B) S3 Intelligent-Tiering only
- C) Manual prod endpoint tweak
- D) Amazon Connect flows
Correct answer: A
Why this is correct: AMT supports Bayesian optimization for efficient search in large spaces.
Why the distractors are incorrect:
- B) S3 Intelligent-Tiering only — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) Manual prod endpoint tweak — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Amazon Connect flows — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 6. Model versions must be approved before production with audit trail.
- A) SageMaker Model Registry
- B) Public S3 ACL
- C) Disable CloudTrail
- D) Amazon Polly lexicons
Correct answer: A
Why this is correct: Model Registry tracks packages and approval status for governed promotion.
Why the distractors are incorrect:
- B) Public S3 ACL — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) Disable CloudTrail — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Amazon Polly lexicons — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 7. Sub-second synchronous predictions 24/7 with autoscaling.
- A) Real-time inference endpoint
- B) Batch Transform only
- C) Glacier retrieval per request
- D) WorkDocs
Correct answer: A
Why this is correct: Real-time endpoints serve low-latency InvokeEndpoint traffic with auto scaling.
Why the distractors are incorrect:
- B) Batch Transform only — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) Glacier retrieval per request — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) WorkDocs — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 8. Monthly scoring of 10M records, no low-latency requirement.
- A) SageMaker Batch Transform
- B) Always-on multi-GPU endpoint
- C) Lambda@Edge per row sync
- D) SES
Correct answer: A
Why this is correct: Batch Transform is cost-effective bulk offline inference from/to S3.
Why the distractors are incorrect:
- B) Always-on multi-GPU endpoint — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) Lambda@Edge per row sync — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) SES — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 9. Payloads up to 1 GB; client polls for results within minutes.
- A) Asynchronous inference endpoint
- B) Serverless endpoint for 1 GB sync
- C) Batch with 1s row SLA
- D) Translate API
Correct answer: A
Why this is correct: Async inference queues large requests and writes output to S3 when complete.
Why the distractors are incorrect:
- B) Serverless endpoint for 1 GB sync — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) Batch with 1s row SLA — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Translate API — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 10. Five small sklearn models on one GPU to save cost.
- A) Multi-Model Endpoint
- B) Five AWS accounts only
- C) Lightsail DB
- D) DeepRacer
Correct answer: A
Why this is correct: MME hosts multiple models from S3 behind one endpoint.
Why the distractors are incorrect:
- B) Five AWS accounts only — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) Lightsail DB — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) DeepRacer — Out of exam scope or violates security and governance requirements.
Question 11. CI/CD on main: build image, test, train, register model.
- A) CodePipeline + CodeBuild + SageMaker Pipelines
- B) Manual SSH
- C) Pinpoint only
- D) No Git
Correct answer: A
Why this is correct: Exam expects CodePipeline/CodeBuild integrated with SageMaker Pipelines.
Why the distractors are incorrect:
- B) Manual SSH — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) Pinpoint only — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) No Git — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 12. Production accuracy drops; input feature distributions shifted.
- A) Data drift — Model Monitor data quality baseline
- B) IAM drift only
- C) Intelligent-Tiering
- D) RI expiration
Correct answer: A
Why this is correct: Data drift is input distribution change detected vs training baseline.
Why the distractors are incorrect:
- B) IAM drift only — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) Intelligent-Tiering — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) RI expiration — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 13. Least-privilege notebook access to one S3 prefix and KMS-encrypted artifacts.
- A) Scoped IAM execution role with kms:Decrypt on the CMK
- B) Root access keys in notebook
- C) s3:* on *
- D) Public bucket ACL
Correct answer: A
Why this is correct: SageMaker assumes IAM roles; scope resources and KMS grants.
Why the distractors are incorrect:
- B) Root access keys in notebook — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) s3:* on * — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Public bucket ACL — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 14. Endpoint only reachable from corporate VPC applications.
- A) VPC config with private subnets and restrictive security groups
- B) Public InvokeEndpoint for all
- C) Remove security groups
- D) Public GitHub model repo
Correct answer: A
Why this is correct: VPC-hosted endpoints keep inference off the public internet.
Why the distractors are incorrect:
- B) Public InvokeEndpoint for all — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) Remove security groups — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Public GitHub model repo — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 15. Right-size inference instance before launch.
- A) SageMaker Inference Recommender
- B) Amazon Translate
- C) DeepRacer
- D) SES bounces
Correct answer: A
Why this is correct: Inference Recommender load-tests candidate instance types.
Why the distractors are incorrect:
- B) Amazon Translate — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) DeepRacer — Out of exam scope or violates security and governance requirements.
- D) SES bounces — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 16. A startup loads 500 GB/day of IoT sensor readings into S3. Analysts query last 7 days by device_id. EMR aggregates hourly. Best format and layout?
- A) Parquet, partitioned by day and device_id
- B) Uncompressed CSV single file
- C) Random JSON blobs without keys
- D) EFS only
Correct answer: A
Why this is correct: Partition pruning on day/device_id minimizes data scanned. Parquet columnar storage suits aggregation. Single CSV/JSON at 500 GB/day is inefficient. EFS is not the primary IoT lake store.
Why the distractors are incorrect:
- B) Uncompressed CSV single file — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) Random JSON blobs without keys — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) EFS only — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 17. You must validate that the `transaction_amount` column has no nulls and stays between 0 and 100000 before a weekly training job. Minimal-code governance approach?
- A) AWS Glue Data Quality rules on the Glue table
- B) Delete the column
- C) Train without validation
- D) Use Amazon Polly
Correct answer: A
Why this is correct: Glue Data Quality integrates with the Data Catalog and can fail jobs when rules break. Deleting features loses signal. Polly is unrelated.
Why the distractors are incorrect:
- B) Delete the column — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) Train without validation — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Use Amazon Polly — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 18. Two teams contribute features with the same record ID but different event times. Feature Store should
- A) use a record identifier and event time for point-in-time correct joins
- B) overwrite all history silently
- C) ignore timestamps
- D) store only in ElastiCache
Correct answer: A
Why this is correct: Feature Store supports event time for point-in-time correctness—critical to avoid label leakage. Silent overwrite breaks reproducibility. ElastiCache lacks offline store semantics.
Why the distractors are incorrect:
- B) overwrite all history silently — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) ignore timestamps — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) store only in ElastiCache — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 19. PII must stay in ap-southeast-1; models train in the same region. Which constraint matters most?
- A) Data residency: S3, KMS keys, and SageMaker resources in ap-southeast-1
- B) Replicate all data to us-east-1 for convenience
- C) Disable encryption
- D) Use public datasets only
Correct answer: A
Why this is correct: Compliance (PII/PHI, residency) requires regional control of storage and keys. Cross-region copies may violate policy. Encryption is required, not disabled.
Why the distractors are incorrect:
- B) Replicate all data to us-east-1 for convenience — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) Disable encryption — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Use public datasets only — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 20. Apache Flink on Managed Service for Apache Flink detects late-arriving events for sessionization. Output features land in S3 as Parquet. This pattern addresses
- A) streaming ETL with event-time windows for real-time features
- B) batch-only reporting with no streams
- C) DNS routing
- D) email campaigns
Correct answer: A
Why this is correct: Flink is in-scope for stream processing; event-time windows handle lateness. DNS/email are unrelated.
Why the distractors are incorrect:
- B) batch-only reporting with no streams — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) DNS routing — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) email campaigns — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 21. Data Wrangler export should feed both SageMaker Processing and scheduled Glue jobs. You should
- A) export reproducible transformation code/recipe into the pipeline
- B) keep transformations only in browser memory
- C) email CSV attachments
- D) avoid version control
Correct answer: A
Why this is correct: Reproducible exports enable MLOps pipelines (exam skill 1.2). Browser-only flows are not repeatable.
Why the distractors are incorrect:
- B) keep transformations only in browser memory — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) email CSV attachments — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) avoid version control — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 22. Selection bias appears because training data only includes users who completed signup. Mitigation includes
- A) revising sampling to include non-completers and measuring bias with Clarify
- B) hiding the issue with more epochs
- C) using larger instance types only
- D) disabling holdout sets
Correct answer: A
Why this is correct: Selection bias is a pre-training concern; Clarify helps measure; sampling fixes the root cause. More epochs or bigger instances do not fix biased data collection.
Why the distractors are incorrect:
- B) hiding the issue with more epochs — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) using larger instance types only — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) disabling holdout sets — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 23. Athena queries on nested JSON in S3 are slow and expensive. First optimization step?
- A) Convert to partitioned Parquet via CTAS or Glue ETL
- B) Increase JSON indentation
- C) Disable S3 versioning
- D) Move to Snowmobile
Correct answer: A
Why this is correct: Columnar Parquet plus partitions reduces bytes scanned—Athena best practice. Snowmobile is bulk migration.
Why the distractors are incorrect:
- B) Increase JSON indentation — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) Disable S3 versioning — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Move to Snowmobile — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 24. Amazon A2I should be used when
- A) human review is needed for low-confidence predictions in a loop with SageMaker
- B) you need only S3 lifecycle rules
- C) training without labels
- D) replacing IAM entirely
Correct answer: A
Why this is correct: Augmented AI (A2I) integrates human reviewers for production ML workflows (e.g., moderation). It does not replace IAM or create labels without a workflow.
Why the distractors are incorrect:
- B) you need only S3 lifecycle rules — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) training without labels — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) replacing IAM entirely — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 25. DynamoDB streams feed Lambda to update online features while S3 holds historical exports. This is
- A) a valid pattern for near-real-time feature updates plus offline history
- B) invalid because DynamoDB cannot stream
- C) only for CloudFront
- D) replaces Feature Store always
Correct answer: A
Why this is correct: Streams + Lambda is a common pattern; Feature Store can still be the canonical store. DynamoDB streams exist. CloudFront is CDN.
Why the distractors are incorrect:
- B) invalid because DynamoDB cannot stream — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) only for CloudFront — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) replaces Feature Store always — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 26. Tokenization for NLP features in a visual pipeline before BERT fine-tuning can be done in
- A) SageMaker Data Wrangler or processing scripts with Hugging Face tokenizers
- B) Route 53
- C) AWS Budgets
- D) Amazon SES
Correct answer: A
Why this is correct: Encoding/tokenization is exam knowledge (task 1.2). Route 53/Budgets/SES are unrelated.
Why the distractors are incorrect:
- B) Route 53 — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) AWS Budgets — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Amazon SES — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 27. Measurement bias occurs when a sensor miscalibrates for one factory. You should
- A) detect distribution differences per factory and recalibrate or exclude bad sensors
- B) merge factories without inspection
- C) increase learning rate
- D) use only accuracy on training set
Correct answer: A
Why this is correct: Measurement bias is a data quality issue addressed in preparation, not hyperparameters alone.
Why the distractors are incorrect:
- B) merge factories without inspection — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) increase learning rate — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) use only accuracy on training set — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 28. Shuffle and split training/validation/test before upload helps
- A) reduce leakage and get unbiased performance estimates
- B) increase S3 API cost only
- C) eliminate need for monitoring
- D) avoid encryption
Correct answer: A
Why this is correct: Proper splitting is task 1.3—prevents overly optimistic metrics and leakage.
Why the distractors are incorrect:
- B) increase S3 API cost only — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) eliminate need for monitoring — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) avoid encryption — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 29. Glue DataBrew is BEST for
- A) visual data preparation without writing Spark code for every transform
- B) hosting real-time GPU inference
- C) domain registration
- D) VPC peering only
Correct answer: A
Why this is correct: DataBrew targets interactive data prep for analytics/ML consumers. It is not an inference or DNS service.
Why the distractors are incorrect:
- B) hosting real-time GPU inference — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) domain registration — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) VPC peering only — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 30. To accelerate repeated reads of a large S3 dataset on training nodes, you can use
- A) FSx for Lustre with lazy loading from S3 or SageMaker fast file mode
- B) Glacier Deep Archive
- C) Public ACL on artifacts
- D) Disable caching entirely always
Correct answer: A
Why this is correct: FSx Lustre and SageMaker input modes address I/O bottlenecks (exam storage guidance).
Why the distractors are incorrect:
- B) Glacier Deep Archive — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) Public ACL on artifacts — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Disable caching entirely always — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 31. Image classification with 1M labeled images and need for transfer learning. Fastest path on SageMaker?
- A) Use a pre-trained model from JumpStart and fine-tune with GPU instances
- B) Train from random weights on CPU only
- C) Use Amazon Translate
- D) Only Batch Transform without training
Correct answer: A
Why this is correct: JumpStart provides pre-trained models; fine-tuning beats training from scratch on large vision data.
Why the distractors are incorrect:
- B) Train from random weights on CPU only — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) Use Amazon Translate — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Only Batch Transform without training — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 32. Random search vs Bayesian optimization for HPO with 100 trials budget on SageMaker AMT
- A) Bayesian explores promising regions faster than naive grid on large spaces
- B) Grid always cheaper
- C) HPO impossible on SageMaker
- D) Only manual tuning allowed
Correct answer: A
Why this is correct: Exam covers Bayesian optimization in AMT for efficient search.
Why the distractors are incorrect:
- B) Grid always cheaper — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) HPO impossible on SageMaker — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Only manual tuning allowed — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 33. Dropout in neural networks primarily acts as
- A) regularization reducing co-adaptation of neurons
- B) data augmentation for images
- C) encryption
- D) S3 replication
Correct answer: A
Why this is correct: Dropout is regularization (task 2.2 knowledge).
Why the distractors are incorrect:
- B) data augmentation for images — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) encryption — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) S3 replication — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 34. AUC-ROC is preferred when
- A) you care about ranking quality across thresholds for imbalanced classes
- B) all classes are perfectly balanced and cost is zero
- C) regression RMSE is the goal
- D) monitoring S3 size
Correct answer: A
Why this is correct: ROC/AUC evaluates classifiers across thresholds—exam metric knowledge.
Why the distractors are incorrect:
- B) all classes are perfectly balanced and cost is zero — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) regression RMSE is the goal — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) monitoring S3 size — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 35. SageMaker Experiments track
- A) parameters, artifacts, and lineage across training runs
- B) only VPC flow logs
- C) employee badges
- D) DNS queries
Correct answer: A
Why this is correct: Experiments support reproducibility (task 2.3).
Why the distractors are incorrect:
- B) only VPC flow logs — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) employee badges — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) DNS queries — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 36. Combining XGBoost, linear model, and neural net predictions via a meta-learner is
- A) stacking ensemble
- B) unsupervised clustering
- C) data drift
- D) Glacier restore
Correct answer: A
Why this is correct: Stacking/ensembling is explicit exam knowledge.
Why the distractors are incorrect:
- B) unsupervised clustering — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) data drift — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Glacier restore — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 37. Early stopping during training saves cost by
- A) halting when validation metric stops improving
- B) running infinite epochs
- C) deleting validation data
- D) disabling checkpoints
Correct answer: A
Why this is correct: Early stopping is a training-time cost/performance method.
Why the distractors are incorrect:
- B) running infinite epochs — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) deleting validation data — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) disabling checkpoints — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 38. For multilingual sentiment at scale without custom training, consider
- A) Amazon Comprehend detect_sentiment
- B) Amazon EBS snapshot
- C) AWS Snowcone per tweet
- D) Amazon MQ
Correct answer: A
Why this is correct: Comprehend is in-scope AI service for NLP tasks.
Why the distractors are incorrect:
- B) Amazon EBS snapshot — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) AWS Snowcone per tweet — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Amazon MQ — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 39. Weight decay (L2) helps by
- A) penalizing large weights to reduce overfitting
- B) increasing model size always
- C) encrypting gradients
- D) replacing validation set
Correct answer: A
Why this is correct: L2 regularization is exam knowledge task 2.2.
Why the distractors are incorrect:
- B) increasing model size always — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) encrypting gradients — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) replacing validation set — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 40. Distributed data parallel training on SageMaker is used when
- A) dataset is large and training time must shrink across many GPUs
- B) dataset fits in 1 MB
- C) inference only
- D) no need for GPUs ever
Correct answer: A
Why this is correct: Distributed training reduces wall-clock time (task 2.2).
Why the distractors are incorrect:
- B) dataset fits in 1 MB — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) inference only — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) no need for GPUs ever — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 41. Clarify SHAP explainability after training helps
- A) interpret feature contributions for stakeholders and debug bias
- B) compress S3 buckets
- C) create IAM users
- D) replace Model Monitor
Correct answer: A
Why this is correct: Clarify explainability is task 2.3; distinct from production monitoring.
Why the distractors are incorrect:
- B) compress S3 buckets — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) create IAM users — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) replace Model Monitor — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 42. Choosing Linear Learner vs XGBoost on tabular data should consider
- A) interpretability, non-linearity, training time, and data size
- B) only logo color
- C) random choice
- D) S3 region name only
Correct answer: A
Why this is correct: Algorithm selection weighs business constraints (task 2.1).
Why the distractors are incorrect:
- B) only logo color — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) random choice — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) S3 region name only — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 43. Catastrophic forgetting in continual learning is mitigated by
- A) rehearsal, regularization, or multi-task training strategies
- B) deleting old data always without review
- C) disabling evaluation
- D) using only Batch Transform
Correct answer: A
Why this is correct: Exam mentions preventing catastrophic forgetting (task 2.2).
Why the distractors are incorrect:
- B) deleting old data always without review — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) disabling evaluation — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) using only Batch Transform — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 44. RMSE is the primary metric for
- A) regression problems
- B) multi-class image tagging only
- C) IAM policy size
- D) S3 object count
Correct answer: A
Why this is correct: RMSE is standard regression metric in guide.
Why the distractors are incorrect:
- B) multi-class image tagging only — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) IAM policy size — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) S3 object count — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 45. Personalize is appropriate when
- A) recommendations for users based on interaction history
- B) OCR of legal PDFs only
- C) network firewall rules
- D) EC2 patching
Correct answer: A
Why this is correct: Amazon Personalize is in-scope ML service for recommendations.
Why the distractors are incorrect:
- B) OCR of legal PDFs only — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) network firewall rules — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) EC2 patching — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 46. Traffic is unpredictable with long idle periods; cold starts up to seconds are acceptable. Choose
- A) SageMaker Serverless Inference
- B) Always-on ml.g5.12xlarge cluster
- C) Snowball Edge
- D) Amazon WorkSpaces
Correct answer: A
Why this is correct: Serverless inference scales to zero and charges per invocation—good for sporadic traffic.
Why the distractors are incorrect:
- B) Always-on ml.g5.12xlarge cluster — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) Snowball Edge — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Amazon WorkSpaces — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 47. CodeDeploy-style linear deployment for endpoints means
- A) gradually increasing traffic to the new variant while monitoring
- B) instant cutover with zero monitoring
- C) deleting old model artifacts
- D) training inside the endpoint
Correct answer: A
Why this is correct: Linear/canary shifts are exam deployment strategies.
Why the distractors are incorrect:
- B) instant cutover with zero monitoring — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) deleting old model artifacts — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) training inside the endpoint — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 48. MWAA (Managed Airflow) is chosen when
- A) existing Airflow DAGs orchestrate ETL and SageMaker jobs
- B) you need a NoSQL database
- C) you want to eliminate orchestration
- D) only for static websites
Correct answer: A
Why this is correct: MWAA is in-scope orchestrator alongside SageMaker Pipelines.
Why the distractors are incorrect:
- B) you need a NoSQL database — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) you want to eliminate orchestration — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) only for static websites — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 49. Step Functions can
- A) coordinate Lambda, SageMaker jobs, and human approval steps in a state machine
- B) only send email
- C) replace S3
- D) train models without data
Correct answer: A
Why this is correct: Step Functions integrates ML workflows in exam scope.
Why the distractors are incorrect:
- B) only send email — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) replace S3 — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) train models without data — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 50. ECR image scanning in CI/CD helps
- A) catch vulnerabilities before deploying inference containers
- B) tune hyperparameters
- C) label images
- D) calculate AUC
Correct answer: A
Why this is correct: Container security is part of deployment best practices.
Why the distractors are incorrect:
- B) tune hyperparameters — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) label images — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) calculate AUC — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 51. Provisioned concurrency on Lambda (when used for ML) reduces
- A) cold start latency for latency-sensitive sporadic models
- B) training time on 8 GPUs
- C) S3 storage cost
- D) need for any IAM roles
Correct answer: A
Why this is correct: Provisioned concurrency addresses cold starts—task 4.2 capacity knowledge.
Why the distractors are incorrect:
- B) training time on 8 GPUs — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) S3 storage cost — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) need for any IAM roles — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 52. Deploying the same model to EKS instead of SageMaker endpoints is chosen when
- A) Kubernetes is the standard platform and custom serving stack exists
- B) you never need containers
- C) data must not be encrypted
- D) you want zero orchestration
Correct answer: A
Why this is correct: EKS/ECS are valid deployment targets per task 3.1.
Why the distractors are incorrect:
- B) you never need containers — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) data must not be encrypted — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) you want zero orchestration — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 53. GitFlow in ML CI/CD typically means
- A) release branches gate production pipeline runs with approvals
- B) no version control
- C) only manual deploys from laptops
- D) shared root credentials in CodeBuild
Correct answer: A
Why this is correct: Git branching models integrate with CodePipeline (task 3.3).
Why the distractors are incorrect:
- B) no version control — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) only manual deploys from laptops — Out of exam scope or violates security and governance requirements.
- D) shared root credentials in CodeBuild — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 54. Integration test in ML pipeline should verify
- A) pipeline stages produce expected artifacts and endpoint health checks pass
- B) only README spelling
- C) employee satisfaction
- D) DNS TTL only
Correct answer: A
Why this is correct: Automated tests include integration/end-to-end per exam.
Why the distractors are incorrect:
- B) only README spelling — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) employee satisfaction — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) DNS TTL only — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 55. Retraining trigger on schema drift should use
- A) EventBridge + Model Monitor/Data Quality alerts → Pipeline execution
- B) random cron yearly only
- C) disabling monitoring
- D) public webhook without auth
Correct answer: A
Why this is correct: Event-driven retraining is core MLOps orchestration.
Why the distractors are incorrect:
- B) random cron yearly only — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) disabling monitoring — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) public webhook without auth — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 56. Neo compiles models for
- A) edge devices with optimized binaries
- B) S3 Glacier
- C) IAM policy editor
- D) Amazon Chime
Correct answer: A
Why this is correct: Neo is edge optimization (task 3.1).
Why the distractors are incorrect:
- B) S3 Glacier — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) IAM policy editor — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Amazon Chime — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 57. Multi-container endpoint hosts
- A) multiple containers in one endpoint (e.g., pre/post processing + model)
- B) only one algorithm globally
- C) data without inference
- D) Route 53 records
Correct answer: A
Why this is correct: Multi-container endpoints are explicit SageMaker capability.
Why the distractors are incorrect:
- B) only one algorithm globally — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) data without inference — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Route 53 records — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 58. On-demand vs provisioned SageMaker inference capacity
- A) on-demand instances scale with traffic; provisioned sets baseline capacity
- B) they are identical
- C) only Snowball provisioned
- D) neither uses EC2
Correct answer: A
Why this is correct: Task 3.2 knowledge: on-demand vs provisioned resources.
Why the distractors are incorrect:
- B) they are identical — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) only Snowball provisioned — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) neither uses EC2 — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 59. CodeBuild in ML CI/CD runs
- A) unit tests, docker build, and artifact packaging
- B) only billing reports
- C) DNS failover
- D) ground truth labeling alone
Correct answer: A
Why this is correct: CodeBuild compiles/tests/builds images—task 3.3.
Why the distractors are incorrect:
- B) only billing reports — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) DNS failover — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) ground truth labeling alone — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 60. Hosting on Lambda for inference limits include
- A) package size, memory, and timeout constraints vs large GPU models
- B) unlimited 30-minute GPU training
- C) no need for IAM
- D) automatic petabyte scaling
Correct answer: A
Why this is correct: Lambda suits small models; large GPU workloads use SageMaker endpoints.
Why the distractors are incorrect:
- B) unlimited 30-minute GPU training — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) no need for IAM — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) automatic petabyte scaling — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 61. Model Monitor bias drift schedule compares
- A) live predictions/features to baseline for fairness metrics over time
- B) S3 bucket names only
- C) employee IDs
- D) CloudFront cache
Correct answer: A
Why this is correct: Bias drift monitoring is SageMaker Clarify/Model Monitor capability.
Why the distractors are incorrect:
- B) S3 bucket names only — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) employee IDs — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) CloudFront cache — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 62. CloudWatch alarm on Invocations and ModelLatency for an endpoint helps
- A) detect performance regressions after deploy
- B) train faster
- C) replace KMS
- D) eliminate VPC
Correct answer: A
Why this is correct: Infrastructure and model latency monitoring are task 4.1/4.2 skills.
Why the distractors are incorrect:
- B) train faster — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) replace KMS — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) eliminate VPC — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 63. Trusted Advisor can recommend
- A) cost optimization and security improvements for ML resources
- B) hyperparameter values
- C) labeling bounding boxes
- D) SQL query plans only
Correct answer: A
Why this is correct: Trusted Advisor is in-scope cost/security tool.
Why the distractors are incorrect:
- B) hyperparameter values — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) labeling bounding boxes — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) SQL query plans only — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 64. SageMaker Savings Plans apply to
- A) committed SageMaker compute usage for cost reduction
- B) S3 Glacier retrieval only
- C) IAM MFA
- D) Route 53 domains
Correct answer: A
Why this is correct: Savings Plans appear in cost optimization task 4.2.
Why the distractors are incorrect:
- B) S3 Glacier retrieval only — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) IAM MFA — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Route 53 domains — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 65. VPC interface endpoint for SageMaker API allows
- A) private connectivity without traversing the public internet
- B) public anonymous access
- C) disabling security groups
- D) removing encryption
Correct answer: A
Why this is correct: Private connectivity is network security best practice.
Why the distractors are incorrect:
- B) public anonymous access — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) disabling security groups — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) removing encryption — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 66. SageMaker Role Manager helps
- A) create least-privilege IAM roles tailored to SageMaker personas
- B) delete all logs
- C) open security groups to 0.0.0.0/0
- D) disable MFA
Correct answer: A
Why this is correct: Role Manager is in-scope IAM helper for ML.
Why the distractors are incorrect:
- B) delete all logs — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) open security groups to 0.0.0.0/0 — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) disable MFA — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 67. X-Ray tracing on inference microservices helps
- A) find latency bottlenecks across Lambda/API Gateway/SageMaker
- B) compute F1 score
- C) label data
- D) tune learning rate
Correct answer: A
Why this is correct: X-Ray is observability for distributed inference paths.
Why the distractors are incorrect:
- B) compute F1 score — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) label data — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) tune learning rate — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 68. A/B testing on SageMaker endpoint variants routes
- A) percentage of traffic to each variant to compare metrics
- B) all traffic blindly to untested model
- C) only DNS queries
- D) training data
Correct answer: A
Why this is correct: A/B on variants is production comparison technique (task 4.1).
Why the distractors are incorrect:
- B) all traffic blindly to untested model — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) only DNS queries — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) training data — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 69. QuickSight dashboards for ML ops can visualize
- A) business KPIs plus CloudWatch-exported operational metrics
- B) only IAM policies
- C) raw binary weights
- D) Git merge conflicts
Correct answer: A
Why this is correct: QuickSight is in-scope for dashboards (task 4.2).
Why the distractors are incorrect:
- B) only IAM policies — Too narrow or contradicts the scenario's scale, latency, or compliance needs.
- C) raw binary weights — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) Git merge conflicts — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 70. Secrets in CodePipeline should
- A) use Secrets Manager or SSM Parameter Store with IAM-scoped access
- B) be hard-coded in buildspec.yml committed to Git
- C) posted in Slack
- D) embedded in public AMIs
Correct answer: A
Why this is correct: CI/CD security best practices forbid hard-coded secrets.
Why the distractors are incorrect:
- B) be hard-coded in buildspec.yml committed to Git — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) posted in Slack — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) embedded in public AMIs — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 71. Service quotas on SageMaker endpoints require
- A) monitoring utilization and requesting limit increases before launch
- B) ignoring limits
- C) using root for all calls
- D) disabling auto scaling always
Correct answer: A
Why this is correct: Capacity troubleshooting includes quotas (task 4.2).
Why the distractors are incorrect:
- B) ignoring limits — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) using root for all calls — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) disabling auto scaling always — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 72. Config rules (AWS Config) can
- A) detect non-compliant SageMaker or S3 settings for governance
- B) train neural networks
- C) perform HPO
- D) replace Model Registry
Correct answer: A
Why this is correct: AWS Config is in-scope governance service.
Why the distractors are incorrect:
- B) train neural networks — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) perform HPO — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) replace Model Registry — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 73. DevOps Guru for ML workloads can
- A) surface operational anomalies using ML on telemetry
- B) label images
- C) compile Neo models
- D) replace CloudTrail
Correct answer: A
Why this is correct: DevOps Guru is in-scope for anomaly detection.
Why the distractors are incorrect:
- B) label images — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) compile Neo models — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) replace CloudTrail — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 74. Encrypting inter-node training traffic in VPC uses
- A) security groups, NACLs, and optional TLS in distributed frameworks
- B) public S3 ACLs
- C) disabling IAM
- D) sharing access keys in notebook
Correct answer: A
Why this is correct: Network controls and encryption protect training clusters.
Why the distractors are incorrect:
- B) public S3 ACLs — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) disabling IAM — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) sharing access keys in notebook — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 75. Budgets alerts at 80% forecasted spend help
- A) proactively control ML experimentation costs
- B) improve model accuracy directly
- C) replace tagging
- D) eliminate CloudWatch
Correct answer: A
Why this is correct: AWS Budgets is explicit cost tool in domain 4.
Why the distractors are incorrect:
- B) improve model accuracy directly — Does not best satisfy the scenario's primary requirement compared to the correct option.
- C) replace tagging — Does not best satisfy the scenario's primary requirement compared to the correct option.
- D) eliminate CloudWatch — Does not best satisfy the scenario's primary requirement compared to the correct option.
Question 76. An ML platform team needs to populate the Glue Data Catalog from S3. Which AWS capability is MOST appropriate?
- A) Use Glue Crawler
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Glue Crawler: enables discoverable schema for ETL and Athena.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 77. An ML platform team needs to migrate on-premises NFS datasets to S3 for training. Which AWS capability is MOST appropriate?
- A) Use Data Sync
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Data Sync: hybrid ingest for large legacy stores.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 78. An ML platform team needs to apply tag-based access control on the data lake. Which AWS capability is MOST appropriate?
- A) Use Lake Formation
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Lake Formation: governed sharing for feature teams.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 79. An ML platform team needs to power vector/text hybrid search features for RAG sidecars. Which AWS capability is MOST appropriate?
- A) Use OpenSearch
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: OpenSearch: in-scope analytics search service.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 80. An ML platform team needs to warehouse joins for feature engineering at scale. Which AWS capability is MOST appropriate?
- A) Use Redshift
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Redshift: structured analytics before SageMaker.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 81. An ML platform team needs to run Spark without managing long-lived clusters. Which AWS capability is MOST appropriate?
- A) Use EMR Serverless
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: EMR Serverless: cost-aware batch feature generation.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 82. An ML platform team needs to notify on-call when Model Monitor violation fires. Which AWS capability is MOST appropriate?
- A) Use SNS
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: SNS: integration for operational alerts.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 83. An ML platform team needs to buffer retraining requests from many producers. Which AWS capability is MOST appropriate?
- A) Use SQS
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: SQS: decouple pipeline triggers.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 84. An ML platform team needs to detect anomalies in operational KPI time series. Which AWS capability is MOST appropriate?
- A) Use Lookout for Metrics
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Lookout for Metrics: managed anomaly detection.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 85. An ML platform team needs to defect detection with custom vision models. Which AWS capability is MOST appropriate?
- A) Use Lookout for Vision
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Lookout for Vision: specialized CV service when not using custom SageMaker.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 86. An ML platform team needs to online fraud scoring with AWS-managed models. Which AWS capability is MOST appropriate?
- A) Use Fraud Detector
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Fraud Detector: pre-built fraud ML service.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 87. An ML platform team needs to extract text from scanned forms for NLP features. Which AWS capability is MOST appropriate?
- A) Use Textract
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Textract: document ML ingest.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 88. An ML platform team needs to speech-to-text for audio feature pipelines. Which AWS capability is MOST appropriate?
- A) Use Transcribe
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Transcribe: audio AI service in scope.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 89. An ML platform team needs to build conversational bots—not primary batch training. Which AWS capability is MOST appropriate?
- A) Use Lex
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Lex: NLP conversational service.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 90. An ML platform team needs to enterprise search over documents. Which AWS capability is MOST appropriate?
- A) Use Kendra
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Kendra: RAG/search complement.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 91. An ML platform team needs to store Python dependencies for reproducible training images. Which AWS capability is MOST appropriate?
- A) Use CodeArtifact
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: CodeArtifact: artifact repo for builds.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 92. An ML platform team needs to declare SageMaker Domain and execution roles as code. Which AWS capability is MOST appropriate?
- A) Use CloudFormation
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: CloudFormation: IaC reproducibility.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 93. An ML platform team needs to define pipeline stacks in TypeScript/Python. Which AWS capability is MOST appropriate?
- A) Use CDK
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: CDK: IaC alternative in scope.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Domain 2: ML Model Development
Question 94. An ML platform team needs to scale endpoint instances on InvocationsPerInstance. Which AWS capability is MOST appropriate?
- A) Use Auto Scaling
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Auto Scaling: elastic inference capacity.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 95. An ML platform team needs to patch and manage notebook instances/hosts. Which AWS capability is MOST appropriate?
- A) Use Systems Manager
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Systems Manager: ops management in scope.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 96. An ML platform team needs to discover PII in S3 before labeling. Which AWS capability is MOST appropriate?
- A) Use Macie
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Macie: sensitive data discovery.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 97. An ML platform team needs to encrypt model artifacts at rest with CMK policies. Which AWS capability is MOST appropriate?
- A) Use KMS
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: KMS: encryption control plane.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 98. An ML platform team needs to audit CreateEndpoint and UpdateEndpoint API calls. Which AWS capability is MOST appropriate?
- A) Use CloudTrail
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: CloudTrail: API audit trail.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 99. An ML platform team needs to analyze SageMaker spend by tag. Which AWS capability is MOST appropriate?
- A) Use Cost Explorer
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Cost Explorer: cost visibility.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 100. An ML platform team needs to recommend smaller inference instances. Which AWS capability is MOST appropriate?
- A) Use Compute Optimizer
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Compute Optimizer: rightsizing recommendations.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 101. An ML platform team needs to schedule data quality against training baseline. Which AWS capability is MOST appropriate?
- A) Use Model Monitor
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Model Monitor: production data drift detection.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 102. An ML platform team needs to capture tensors to detect vanishing gradients. Which AWS capability is MOST appropriate?
- A) Use Debugger
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Debugger: training convergence debug.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 103. An ML platform team needs to run Spark/scikit preprocessing at scale. Which AWS capability is MOST appropriate?
- A) Use Processing Job
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Processing Job: managed feature engineering jobs.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 104. An ML platform team needs to run distributed GPU script mode. Which AWS capability is MOST appropriate?
- A) Use Training Job
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Training Job: core training primitive.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 105. An ML platform team needs to DAG of Processing/Training/Register steps. Which AWS capability is MOST appropriate?
- A) Use Pipeline
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Pipeline: native ML workflow orchestration.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 106. An ML platform team needs to deploy pre-trained models quickly. Which AWS capability is MOST appropriate?
- A) Use JumpStart
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: JumpStart: accelerator for common architectures.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 107. An ML platform team needs to invoke foundation models without self-hosting weights. Which AWS capability is MOST appropriate?
- A) Use Bedrock
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Bedrock: managed FM inference.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 108. An ML platform team needs to face/object APIs without custom training. Which AWS capability is MOST appropriate?
- A) Use Rekognition
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Rekognition: vision AI service.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 109. An ML platform team needs to PHI-aware NLP on clinical text. Which AWS capability is MOST appropriate?
- A) Use Comprehend Medical
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Comprehend Medical: healthcare NLP service.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 110. An ML platform team needs to attach accelerator to CPU instances (legacy pattern). Which AWS capability is MOST appropriate?
- A) Use Elastic Inference
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Elastic Inference: cost inference acceleration where applicable.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 111. An ML platform team needs to run containerized batch scoring jobs at scale. Which AWS capability is MOST appropriate?
- A) Use Batch
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Batch: non-SageMaker batch compute option.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 112. An ML platform team needs to run custom inference containers orchestrated by ECS. Which AWS capability is MOST appropriate?
- A) Use ECS
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: ECS: container deployment target.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 113. An ML platform team needs to expose Lambda/SageMaker behind REST API. Which AWS capability is MOST appropriate?
- A) Use API Gateway
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: API Gateway: edge API for ML microservices.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 114. An ML platform team needs to lightweight inference for sporadic low-volume models. Which AWS capability is MOST appropriate?
- A) Use Lambda
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Lambda: serverless inference pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 115. An ML platform team needs to react to S3 Object Created for pipelines. Which AWS capability is MOST appropriate?
- A) Use EventBridge
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: EventBridge: event-driven MLOps.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 116. An ML platform team needs to visual profiling and recipes without Spark code. Which AWS capability is MOST appropriate?
- A) Use DataBrew
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: DataBrew: interactive data prep.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 117. An ML platform team needs to managed labeling with quality controls. Which AWS capability is MOST appropriate?
- A) Use Ground Truth
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Ground Truth: dataset labeling at scale.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 118. An ML platform team needs to online/offline feature consistency. Which AWS capability is MOST appropriate?
- A) Use Feature Store
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Feature Store: feature management service.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 119. An ML platform team needs to SHAP and bias reports. Which AWS capability is MOST appropriate?
- A) Use Clarify
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Clarify: explainability and fairness insights.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 120. An ML platform team needs to governed model promotion. Which AWS capability is MOST appropriate?
- A) Use Model Registry
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Model Registry: versioning and approvals.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 121. An ML platform team needs to compile models for edge hardware. Which AWS capability is MOST appropriate?
- A) Use Neo
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Neo: edge optimization.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 122. An ML platform team needs to benchmark latency/throughput across instances. Which AWS capability is MOST appropriate?
- A) Use Inference Recommender
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Inference Recommender: sizing guidance.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 123. An ML platform team needs to discount training with checkpoint resume. Which AWS capability is MOST appropriate?
- A) Use Spot Training
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Spot Training: cost-optimized training.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 124. An ML platform team needs to compare new model on copied traffic. Which AWS capability is MOST appropriate?
- A) Use Shadow variant
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Shadow variant: safe production evaluation.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 125. An ML platform team needs to shift small traffic % then increase. Which AWS capability is MOST appropriate?
- A) Use Canary deployment
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Canary deployment: risk-managed rollout.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 126. An ML platform team needs to parallel stacks then switch traffic. Which AWS capability is MOST appropriate?
- A) Use Blue/green deployment
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Blue/green deployment: rollback-friendly release.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 127. An ML platform team needs to trigger CodePipeline from repo webhook. Which AWS capability is MOST appropriate?
- A) Use GitHub Actions (via CodeSuite)
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: GitHub Actions (via CodeSuite): CI integration pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 128. An ML platform team needs to define CodeBuild phases for test and docker build. Which AWS capability is MOST appropriate?
- A) Use Buildspec
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Buildspec: CI build definition.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 129. An ML platform team needs to manual gate in CodePipeline before prod deploy. Which AWS capability is MOST appropriate?
- A) Use Approval stage
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Approval stage: human-in-the-loop governance.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 130. An ML platform team needs to allocate ML costs to cost centers. Which AWS capability is MOST appropriate?
- A) Use Tagging strategy
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Tagging strategy: financial governance.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 131. An ML platform team needs to alert when forecast exceeds monthly ML cap. Which AWS capability is MOST appropriate?
- A) Use Budgets
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Budgets: proactive spend control.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 132. An ML platform team needs to dashboard model KPIs for business users. Which AWS capability is MOST appropriate?
- A) Use QuickSight
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: QuickSight: BI visualization.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 133. An ML platform team needs to query inference logs for errors. Which AWS capability is MOST appropriate?
- A) Use CloudWatch Logs Insights
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: CloudWatch Logs Insights: log analytics.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 134. An ML platform team needs to trace latency across microservices. Which AWS capability is MOST appropriate?
- A) Use X-Ray
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: X-Ray: distributed tracing.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 135. An ML platform team needs to private S3/SageMaker access without NAT. Which AWS capability is MOST appropriate?
- A) Use VPC endpoints
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: VPC endpoints: network isolation.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 136. An ML platform team needs to restrict endpoint ENI to app subnet CIDR. Which AWS capability is MOST appropriate?
- A) Use Security group
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Security group: network ACL at instance level.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 137. An ML platform team needs to restrict SageMaker actions by vpc/tag. Which AWS capability is MOST appropriate?
- A) Use IAM condition keys
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: IAM condition keys: fine-grained authorization.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 138. An ML platform team needs to deny non-TLS S3 access for artifacts. Which AWS capability is MOST appropriate?
- A) Use Bucket policy
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Bucket policy: data perimeter control.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 139. An ML platform team needs to customer-managed keys for compliance. Which AWS capability is MOST appropriate?
- A) Use SSE-S3 vs SSE-KMS
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: SSE-S3 vs SSE-KMS: KMS for key policy control.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 140. An ML platform team needs to allow central tooling account to deploy models. Which AWS capability is MOST appropriate?
- A) Use Cross-account role
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Cross-account role: enterprise access pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 141. An ML platform team needs to collaborative IDE for experiments. Which AWS capability is MOST appropriate?
- A) Use SageMaker Studio
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: SageMaker Studio: unified ML IDE.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 142. An ML platform team needs to automate notebook setup scripts. Which AWS capability is MOST appropriate?
- A) Use Lifecycle config
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Lifecycle config: repeatable dev environments.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 143. An ML platform team needs to stream S3 data without full download. Which AWS capability is MOST appropriate?
- A) Use Pipe input mode
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Pipe input mode: training I/O optimization.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 144. An ML platform team needs to copy dataset to training volume. Which AWS capability is MOST appropriate?
- A) Use File input mode
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: File input mode: when repeated local reads help.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 145. An ML platform team needs to optimized mount for large datasets. Which AWS capability is MOST appropriate?
- A) Use FastFile mode
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: FastFile mode: reduced startup for big data.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 146. An ML platform team needs to batch size affects memory and convergence. Which AWS capability is MOST appropriate?
- A) Use Hyperparameter
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Hyperparameter: tuning impacts training stability.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 147. An ML platform team needs to epoch is full pass; step is one batch update. Which AWS capability is MOST appropriate?
- A) Use Epoch vs step
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Epoch vs step: training vocabulary.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 148. An ML platform team needs to combine models to reduce variance. Which AWS capability is MOST appropriate?
- A) Use Ensembling
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Ensembling: performance technique.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 149. An ML platform team needs to remove weights to shrink model size. Which AWS capability is MOST appropriate?
- A) Use Pruning
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Pruning: compression technique in scope.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 150. An ML platform team needs to visualize CNN focus for debugging. Which AWS capability is MOST appropriate?
- A) Use Heat map
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Heat map: evaluation visualization.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 151. An ML platform team needs to see per-class errors. Which AWS capability is MOST appropriate?
- A) Use Confusion matrix
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Confusion matrix: classification evaluation.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 152. An ML platform team needs to tune threshold for fraud use cases. Which AWS capability is MOST appropriate?
- A) Use Precision/recall tradeoff
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Precision/recall tradeoff: metric interpretation.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 153. An ML platform team needs to high error on train and validation. Which AWS capability is MOST appropriate?
- A) Use Underfitting
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Underfitting: needs more capacity or features.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 154. An ML platform team needs to low train error, high validation error. Which AWS capability is MOST appropriate?
- A) Use Overfitting
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Overfitting: needs regularization or more data.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 155. An ML platform team needs to P(Y|X) changes over time. Which AWS capability is MOST appropriate?
- A) Use Concept drift
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Concept drift: monitor model quality metrics.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 156. An ML platform team needs to features contain future information. Which AWS capability is MOST appropriate?
- A) Use Label leakage
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Label leakage: prevent with point-in-time joins.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 157. An ML platform team needs to categorical to binary columns. Which AWS capability is MOST appropriate?
- A) Use One-hot encoding
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: One-hot encoding: encoding technique task 1.2.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 158. An ML platform team needs to discretize continuous variables. Which AWS capability is MOST appropriate?
- A) Use Binning
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Binning: feature engineering technique.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 159. An ML platform team needs to stabilize skewed numeric features. Which AWS capability is MOST appropriate?
- A) Use Log transform
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Log transform: feature engineering technique.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 160. An ML platform team needs to efficient SageMaker built-in ingest. Which AWS capability is MOST appropriate?
- A) Use RecordIO protobuf
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: RecordIO protobuf: format for algorithms.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 161. An ML platform team needs to columnar alternative to Parquet. Which AWS capability is MOST appropriate?
- A) Use ORC
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: ORC: valid lake format.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 162. An ML platform team needs to schema evolution friendly row format. Which AWS capability is MOST appropriate?
- A) Use Avro
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Avro: common in Kafka sinks.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 163. An ML platform team needs to parallelism unit for streams. Which AWS capability is MOST appropriate?
- A) Use Kinesis shard
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Kinesis shard: streaming scalability concept.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 164. An ML platform team needs to batch small records to S3 objects. Which AWS capability is MOST appropriate?
- A) Use Firehose buffering
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Firehose buffering: efficient land in lake.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 165. An ML platform team needs to crowd labeling integrated via Ground Truth. Which AWS capability is MOST appropriate?
- A) Use Mechanical Turk
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Mechanical Turk: labeling workforce option.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 166. An ML platform team needs to human review loop for low-confidence predictions. Which AWS capability is MOST appropriate?
- A) Use A2I
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: A2I: human-in-the-loop production.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 167. An ML platform team needs to real-time recommendations API. Which AWS capability is MOST appropriate?
- A) Use Personalize
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Personalize: recommendation service.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 168. An ML platform team needs to text-to-speech—not training labels. Which AWS capability is MOST appropriate?
- A) Use Polly
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Polly: speech synthesis service.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 169. An ML platform team needs to language translation API. Which AWS capability is MOST appropriate?
- A) Use Translate
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Translate: NLP service selection.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 170. An ML platform team needs to FHIR data store for healthcare analytics. Which AWS capability is MOST appropriate?
- A) Use HealthLake
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: HealthLake: health data service.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 171. An ML platform team needs to ML-powered code review for quality. Which AWS capability is MOST appropriate?
- A) Use CodeGuru
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: CodeGuru: dev quality tool in ML category.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 172. An ML platform team needs to manage edge deployments (where used). Which AWS capability is MOST appropriate?
- A) Use SageMaker Edge Manager
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: SageMaker Edge Manager: edge lifecycle complement to Neo.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 173. An ML platform team needs to sidecar preprocessing container. Which AWS capability is MOST appropriate?
- A) Use Multi-container endpoint
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Multi-container endpoint: deploy pre/post with model.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 174. An ML platform team needs to SNS when inference output ready. Which AWS capability is MOST appropriate?
- A) Use Async inference notification
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Async inference notification: client notification pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 175. An ML platform team needs to scale to zero with cold starts. Which AWS capability is MOST appropriate?
- A) Use Serverless inference
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Serverless inference: intermittent traffic pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 176. An ML platform team needs to A/B traffic split on endpoint. Which AWS capability is MOST appropriate?
- A) Use Real-time multi-variant
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Real-time multi-variant: production experimentation.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 177. An ML platform team needs to list S3 objects to score. Which AWS capability is MOST appropriate?
- A) Use Batch Transform manifest
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Batch Transform manifest: batch input format.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 178. An ML platform team needs to organize versions in registry. Which AWS capability is MOST appropriate?
- A) Use Model package group
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Model package group: registry structure.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 179. An ML platform team needs to pass dynamic config between steps. Which AWS capability is MOST appropriate?
- A) Use Pipeline parameter
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Pipeline parameter: parameterized workflows.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 180. An ML platform team needs to reuse outputs when inputs unchanged. Which AWS capability is MOST appropriate?
- A) Use CacheConfig on pipeline step
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: CacheConfig on pipeline step: pipeline cost optimization.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Domain 3: Deployment and Orchestration
Question 181. An ML platform team needs to handle transient training failures. Which AWS capability is MOST appropriate?
- A) Use Retry policy
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Retry policy: resilient orchestration.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 182. An ML platform team needs to checkpoint resume from S3. Which AWS capability is MOST appropriate?
- A) Use Spot failure
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Spot failure: fault-tolerant spot training.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 183. An ML platform team needs to keep data plane private. Which AWS capability is MOST appropriate?
- A) Use VPC-only S3 gateway endpoint
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: VPC-only S3 gateway endpoint: network cost/privacy.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 184. An ML platform team needs to allow private subnet outbound for pip install. Which AWS capability is MOST appropriate?
- A) Use NAT gateway
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: NAT gateway: egress for package installs.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 185. An ML platform team needs to private connectivity to SageMaker API. Which AWS capability is MOST appropriate?
- A) Use Interface endpoint
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Interface endpoint: privateLink pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 186. An ML platform team needs to faster cross-region uploads. Which AWS capability is MOST appropriate?
- A) Use S3 Transfer Acceleration
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: S3 Transfer Acceleration: distant client ingest.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 187. An ML platform team needs to transition old artifacts to IA/Glacier. Which AWS capability is MOST appropriate?
- A) Use S3 lifecycle
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: S3 lifecycle: artifact cost control.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 188. An ML platform team needs to low-latency key-value online features. Which AWS capability is MOST appropriate?
- A) Use DynamoDB feature
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: DynamoDB feature: alternative online store pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 189. An ML platform team needs to caching layer—not full feature store. Which AWS capability is MOST appropriate?
- A) Use ElastiCache
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: ElastiCache: cache vs feature store distinction.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 190. An ML platform team needs to transactional source—not primary lake. Which AWS capability is MOST appropriate?
- A) Use RDS
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: RDS: OLTP vs analytics separation.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 191. An ML platform team needs to document source for features. Which AWS capability is MOST appropriate?
- A) Use DocumentDB
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: DocumentDB: NoSQL ingest option.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 192. An ML platform team needs to graph relationships as features. Which AWS capability is MOST appropriate?
- A) Use Neptune
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Neptune: graph ML ingest.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 193. An ML platform team needs to create Parquet tables from queries. Which AWS capability is MOST appropriate?
- A) Use Athena CTAS
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Athena CTAS: lake transformation pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 194. An ML platform team needs to incremental ETL processing. Which AWS capability is MOST appropriate?
- A) Use Glue bookmark
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Glue bookmark: incremental pipeline state.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 195. An ML platform team needs to distributed transform for features. Which AWS capability is MOST appropriate?
- A) Use Spark on Glue
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Spark on Glue: serverless Spark option.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 196. An ML platform team needs to long-running Spark when always-on needed. Which AWS capability is MOST appropriate?
- A) Use EMR cluster
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: EMR cluster: cluster tradeoff vs Serverless.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 197. An ML platform team needs to fail job on null rate threshold. Which AWS capability is MOST appropriate?
- A) Use Data Quality ruleset
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Data Quality ruleset: quality gate before training.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 198. An ML platform team needs to difference in proportions of labels across groups. Which AWS capability is MOST appropriate?
- A) Use Clarify DPL
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Clarify DPL: bias metric name.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 199. An ML platform team needs to class imbalance measure. Which AWS capability is MOST appropriate?
- A) Use Clarify CI
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Clarify CI: bias metric name.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 200. An ML platform team needs to balance rare classes. Which AWS capability is MOST appropriate?
- A) Use Synthetic data
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Synthetic data: mitigate imbalance.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 201. An ML platform team needs to expand training images via transforms. Which AWS capability is MOST appropriate?
- A) Use Augmentation
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Augmentation: vision regularization/data volume.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 202. An ML platform team needs to unbiased estimate of generalization. Which AWS capability is MOST appropriate?
- A) Use Train/val/test split
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Train/val/test split: fundamental modeling practice.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 203. An ML platform team needs to reproducible experiments. Which AWS capability is MOST appropriate?
- A) Use Random seed
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Random seed: experiment tracking practice.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 204. An ML platform team needs to bring custom training script. Which AWS capability is MOST appropriate?
- A) Use Script mode
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Script mode: custom framework training.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 205. An ML platform team needs to custom inference container from ECR. Which AWS capability is MOST appropriate?
- A) Use BYOC inference
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: BYOC inference: custom serving stack.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 206. An ML platform team needs to start SageMaker training job via SDK. Which AWS capability is MOST appropriate?
- A) Use Estimator.fit
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Estimator.fit: SDK training invocation.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 207. An ML platform team needs to high-level deploy from trained estimator. Which AWS capability is MOST appropriate?
- A) Use Predictor
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Predictor: SDK deployment helper.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 208. An ML platform team needs to objective metric for AMT jobs. Which AWS capability is MOST appropriate?
- A) Use HPO metric
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: HPO metric: tuning optimization target.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 209. An ML platform team needs to stop unpromising training jobs in HPO. Which AWS capability is MOST appropriate?
- A) Use Early stopping rule
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Early stopping rule: save tuning budget.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 210. An ML platform team needs to use prior tuning job results. Which AWS capability is MOST appropriate?
- A) Use Warm start tuning
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Warm start tuning: accelerate HPO.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 211. An ML platform team needs to use Spot instances for Training jobs. Which AWS capability is MOST appropriate?
- A) Use Managed Spot Training
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Managed Spot Training: training cost optimization.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 212. An ML platform team needs to optimize models for faster training/inference. Which AWS capability is MOST appropriate?
- A) Use Training Compiler
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Training Compiler: compilation optimization.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 213. An ML platform team needs to split large models across GPUs. Which AWS capability is MOST appropriate?
- A) Use Model parallelism
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Model parallelism: large model training.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 214. An ML platform team needs to split batches across replicas. Which AWS capability is MOST appropriate?
- A) Use Data parallelism
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Data parallelism: scale training throughput.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 215. An ML platform team needs to stream data from S3 during training. Which AWS capability is MOST appropriate?
- A) Use Pipe mode
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Pipe mode: I/O optimization.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 216. An ML platform team needs to list of training samples in S3. Which AWS capability is MOST appropriate?
- A) Use Manifest file
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Manifest file: input definition pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 217. An ML platform team needs to bounding boxes for object detection. Which AWS capability is MOST appropriate?
- A) Use Augmented manifest
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Augmented manifest: structured label format.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 218. An ML platform team needs to custom Docker training image. Which AWS capability is MOST appropriate?
- A) Use Bring your own algorithm
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Bring your own algorithm: maximum training flexibility.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 219. An ML platform team needs to batch multiple rows per request on endpoint. Which AWS capability is MOST appropriate?
- A) Use Multi-record batch inference
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Multi-record batch inference: throughput optimization.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 220. An ML platform team needs to TargetTrackingScaling on CPU or custom metric. Which AWS capability is MOST appropriate?
- A) Use Auto scaling target
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Auto scaling target: scaling policy type.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 221. An ML platform team needs to prevent flapping when traffic drops. Which AWS capability is MOST appropriate?
- A) Use Scale-in cooldown
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Scale-in cooldown: stability of autoscaling.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 222. An ML platform team needs to reserved capacity pattern (where applicable). Which AWS capability is MOST appropriate?
- A) Use Provisioned throughput mode
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Provisioned throughput mode: capacity planning concept.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 223. An ML platform team needs to no user impact from shadow responses. Which AWS capability is MOST appropriate?
- A) Use Shadow production variant
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Shadow production variant: safe comparison.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 224. An ML platform team needs to traffic split percentage. Which AWS capability is MOST appropriate?
- A) Use Production variant weight
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Production variant weight: A/B configuration.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 225. An ML platform team needs to immutable definition of variants and instances. Which AWS capability is MOST appropriate?
- A) Use Endpoint config
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Endpoint config: deployment artifact.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 226. An ML platform team needs to move endpoint to new config with rollout. Which AWS capability is MOST appropriate?
- A) Use UpdateEndpoint
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: UpdateEndpoint: deployment API.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 227. An ML platform team needs to revert to previous endpoint config on alarms. Which AWS capability is MOST appropriate?
- A) Use Rollback
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Rollback: operational safety.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 228. An ML platform team needs to invoke endpoint with sample payload. Which AWS capability is MOST appropriate?
- A) Use Integration test in pipeline
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Integration test in pipeline: automated quality gate.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 229. An ML platform team needs to validate data transform functions. Which AWS capability is MOST appropriate?
- A) Use Unit test in CodeBuild
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Unit test in CodeBuild: shift-left quality.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 230. An ML platform team needs to production pipeline triggered from release/*. Which AWS capability is MOST appropriate?
- A) Use Gitflow release branch
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Gitflow release branch: branching strategy.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 231. An ML platform team needs to encrypt CodePipeline artifacts with KMS. Which AWS capability is MOST appropriate?
- A) Use Artifact encryption
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Artifact encryption: CI/CD security.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 232. An ML platform team needs to CodeBuild role scoped to ECR and S3 prefixes. Which AWS capability is MOST appropriate?
- A) Use Least privilege build role
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Least privilege build role: pipeline IAM best practice.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 233. An ML platform team needs to page ops when inference errors spike. Which AWS capability is MOST appropriate?
- A) Use CloudWatch alarm on 5xx
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: CloudWatch alarm on 5xx: operational monitoring.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 234. An ML platform team needs to publish business KPI from inference container. Which AWS capability is MOST appropriate?
- A) Use Custom metric
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Custom metric: application-level monitoring.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 235. An ML platform team needs to control storage cost of verbose inference logs. Which AWS capability is MOST appropriate?
- A) Use Log group retention
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Log group retention: log cost management.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 236. An ML platform team needs to mask or exclude from CloudWatch. Which AWS capability is MOST appropriate?
- A) Use PII in logs
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: PII in logs: compliance in observability.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 237. An ML platform team needs to prevent use of unapproved regions for ML. Which AWS capability is MOST appropriate?
- A) Use Organizations SCP
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Organizations SCP: guardrail at org level.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 238. An ML platform team needs to approved SageMaker products for teams. Which AWS capability is MOST appropriate?
- A) Use Service Catalog
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Service Catalog: governed self-service.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 239. An ML platform team needs to Slack notifications from CloudWatch alarms. Which AWS capability is MOST appropriate?
- A) Use Chatbot
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Chatbot: ops integration.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 240. An ML platform team needs to underutilized SageMaker endpoints. Which AWS capability is MOST appropriate?
- A) Use Trusted Advisor check
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Trusted Advisor check: waste identification.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 241. An ML platform team needs to commit to consistent SageMaker compute. Which AWS capability is MOST appropriate?
- A) Use Savings Plans
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Savings Plans: discount for steady workloads.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 242. An ML platform team needs to reserve instances for always-on endpoints. Which AWS capability is MOST appropriate?
- A) Use RI
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: RI: predictable baseline cost.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 243. An ML platform team needs to generally avoided for steady SLA real-time. Which AWS capability is MOST appropriate?
- A) Use Spot for inference
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Spot for inference: Spot better for fault-tolerant batch/train.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 244. An ML platform team needs to stop idle Studio apps to save cost. Which AWS capability is MOST appropriate?
- A) Use Rightsizing notebook
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Rightsizing notebook: dev environment cost control.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 245. An ML platform team needs to dev/stage/prod isolation for ML. Which AWS capability is MOST appropriate?
- A) Use Separate accounts
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Separate accounts: blast radius reduction.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 246. An ML platform team needs to DR for model artifacts—watch residency. Which AWS capability is MOST appropriate?
- A) Use Cross-region replication
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Cross-region replication: BC/DR vs compliance tradeoff.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 247. An ML platform team needs to document intended use and limitations. Which AWS capability is MOST appropriate?
- A) Use Model card
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Model card: governance documentation practice.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 248. An ML platform team needs to diverse labelers in Ground Truth. Which AWS capability is MOST appropriate?
- A) Use Human review bias
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Human review bias: labeling quality practice.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 249. An ML platform team needs to measure labeling quality. Which AWS capability is MOST appropriate?
- A) Use Inter-annotator agreement
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Inter-annotator agreement: Ground Truth quality mechanism.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 250. An ML platform team needs to label uncertain samples first. Which AWS capability is MOST appropriate?
- A) Use Active learning
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Active learning: efficient labeling strategy.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 251. An ML platform team needs to fine-tune vs train from scratch. Which AWS capability is MOST appropriate?
- A) Use Transfer learning
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Transfer learning: faster convergence with less data.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 252. An ML platform team needs to large pre-trained model adapted downstream. Which AWS capability is MOST appropriate?
- A) Use Foundation model
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Foundation model: modern NLP/CV approach.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 253. An ML platform team needs to optimize inputs to FM APIs. Which AWS capability is MOST appropriate?
- A) Use Prompt engineering
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Prompt engineering: Bedrock usage pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Domain 4: Monitoring, Maintenance, and Security
Question 254. An ML platform team needs to retrieve context then generate answer. Which AWS capability is MOST appropriate?
- A) Use RAG
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: RAG: reduce hallucination pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 255. An ML platform team needs to scheduled Batch Transform after ETL. Which AWS capability is MOST appropriate?
- A) Use Batch inference pipeline
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Batch inference pipeline: scheduled scoring pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 256. An ML platform team needs to Kinesis + SageMaker Processing/Endpoints. Which AWS capability is MOST appropriate?
- A) Use Streaming inference
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Streaming inference: near-real-time pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 257. An ML platform team needs to test feature code on PR in CodeBuild. Which AWS capability is MOST appropriate?
- A) Use Feature pipeline CI
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Feature pipeline CI: feature store governance.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 258. An ML platform team needs to schema expectations between teams. Which AWS capability is MOST appropriate?
- A) Use Data contract
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Data contract: prevent breaking changes.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 259. An ML platform team needs to Glue schema version for Avro/JSON. Which AWS capability is MOST appropriate?
- A) Use Schema registry
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Schema registry: evolution control.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 260. An ML platform team needs to tokenize before labeling. Which AWS capability is MOST appropriate?
- A) Use PII masking
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: PII masking: privacy technique.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 261. An ML platform team needs to add noise for privacy (awareness). Which AWS capability is MOST appropriate?
- A) Use Differential privacy
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Differential privacy: advanced privacy concept.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 262. An ML platform team needs to Clarify post-training bias metrics. Which AWS capability is MOST appropriate?
- A) Use Model bias post-training
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Model bias post-training: fairness after training.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 263. An ML platform team needs to SHAP for regulated decisions. Which AWS capability is MOST appropriate?
- A) Use Explainability threshold
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Explainability threshold: interpretability requirement.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 264. An ML platform team needs to awareness for security testing models. Which AWS capability is MOST appropriate?
- A) Use Adversarial robustness
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Adversarial robustness: ML security awareness.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 265. An ML platform team needs to encrypt data in transit to InvokeEndpoint. Which AWS capability is MOST appropriate?
- A) Use Endpoint TLS
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Endpoint TLS: encryption in transit.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 266. An ML platform team needs to mutual TLS for service-to-service. Which AWS capability is MOST appropriate?
- A) Use mTLS
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: mTLS: advanced auth pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 267. An ML platform team needs to consumer VPC accesses SageMaker privately. Which AWS capability is MOST appropriate?
- A) Use PrivateLink
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: PrivateLink: private connectivity.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 268. An ML platform team needs to pipeline must pass SageMaker execution role. Which AWS capability is MOST appropriate?
- A) Use IAM PassRole
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: IAM PassRole: common permission error.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 269. An ML platform team needs to S3 bucket policy allowing SageMaker role. Which AWS capability is MOST appropriate?
- A) Use Resource-based policy
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Resource-based policy: cross-service access.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 270. An ML platform team needs to restrict S3 access to VPC endpoint. Which AWS capability is MOST appropriate?
- A) Use Condition aws:SourceVpc
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Condition aws:SourceVpc: data exfiltration prevention.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 271. An ML platform team needs to Registry + Pipeline execution ARN. Which AWS capability is MOST appropriate?
- A) Use Audit model lineage
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Audit model lineage: traceability for compliance.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 272. An ML platform team needs to EventBridge cron starts Pipeline. Which AWS capability is MOST appropriate?
- A) Use Retrain on schedule
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Retrain on schedule: scheduled refresh.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 273. An ML platform team needs to CloudWatch alarm → Lambda → Pipeline. Which AWS capability is MOST appropriate?
- A) Use Retrain on drift
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Retrain on drift: conditional refresh.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 274. An ML platform team needs to challenger variant gets minority traffic. Which AWS capability is MOST appropriate?
- A) Use Champion/challenger
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Champion/challenger: continuous evaluation pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 275. An ML platform team needs to never used during HPO to avoid leakage. Which AWS capability is MOST appropriate?
- A) Use Holdout set
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Holdout set: honest final evaluation.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 276. An ML platform team needs to k-fold on training for robust metrics. Which AWS capability is MOST appropriate?
- A) Use Cross-validation
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Cross-validation: evaluation technique.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 277. An ML platform team needs to preserve class ratios in splits. Which AWS capability is MOST appropriate?
- A) Use Stratified split
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Stratified split: imbalanced data practice.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 278. An ML platform team needs to synthetic minority oversampling technique. Which AWS capability is MOST appropriate?
- A) Use SMOTE
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: SMOTE: imbalance handling option.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 279. An ML platform team needs to penalize misclassification of rare class. Which AWS capability is MOST appropriate?
- A) Use Class weights
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Class weights: algorithm-level imbalance handling.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 280. An ML platform team needs to adjust decision threshold on validation. Which AWS capability is MOST appropriate?
- A) Use Threshold tuning
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Threshold tuning: optimize precision/recall.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 281. An ML platform team needs to align predicted probabilities with outcomes. Which AWS capability is MOST appropriate?
- A) Use Calibration
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Calibration: probability quality.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 282. An ML platform team needs to choose appropriate loss and metrics. Which AWS capability is MOST appropriate?
- A) Use Multiclass vs multilabel
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Multiclass vs multilabel: problem framing.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 283. An ML platform team needs to mAP vs simple accuracy. Which AWS capability is MOST appropriate?
- A) Use Object detection metric
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Object detection metric: CV evaluation awareness.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 284. An ML platform team needs to pixel-level masks—specialized algorithms. Which AWS capability is MOST appropriate?
- A) Use Segmentation
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Segmentation: CV task type.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 285. An ML platform team needs to forecasting algorithms (DeepAR etc.). Which AWS capability is MOST appropriate?
- A) Use Time series
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Time series: SageMaker forecasting domain.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 286. An ML platform team needs to Personalize campaign for new users. Which AWS capability is MOST appropriate?
- A) Use Cold start recommendation
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Cold start recommendation: recommendation feature.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 287. An ML platform team needs to online store vs offline store latency. Which AWS capability is MOST appropriate?
- A) Use Real-time vs batch features
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Real-time vs batch features: Feature Store architecture.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 288. An ML platform team needs to join features as of event timestamp. Which AWS capability is MOST appropriate?
- A) Use Point-in-time query
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Point-in-time query: prevent leakage.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 289. An ML platform team needs to logical collection in Feature Store. Which AWS capability is MOST appropriate?
- A) Use Feature group
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Feature group: Feature Store primitive.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 290. An ML platform team needs to historical features for training. Which AWS capability is MOST appropriate?
- A) Use Offline store S3
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Offline store S3: batch training data source.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 291. An ML platform team needs to low-latency serving layer. Which AWS capability is MOST appropriate?
- A) Use Online store DynamoDB
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Online store DynamoDB: inference feature retrieval.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 292. An ML platform team needs to expire stale online features. Which AWS capability is MOST appropriate?
- A) Use TTL on features
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: TTL on features: freshness control.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 293. An ML platform team needs to populate historical features. Which AWS capability is MOST appropriate?
- A) Use Backfill pipeline
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Backfill pipeline: Feature Store operations.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 294. An ML platform team needs to explore before production pipeline. Which AWS capability is MOST appropriate?
- A) Use Ingestion notebook
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Ingestion notebook: iterative then operationalize.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 295. An ML platform team needs to non-secret config for pipelines. Which AWS capability is MOST appropriate?
- A) Use Parameter Store
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Parameter Store: configuration management.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 296. An ML platform team needs to rotate DB creds without downtime. Which AWS capability is MOST appropriate?
- A) Use Secrets rotation
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Secrets rotation: Secrets Manager capability.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 297. An ML platform team needs to speed docker layer rebuilds. Which AWS capability is MOST appropriate?
- A) Use Build cache in CodeBuild
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Build cache in CodeBuild: CI performance.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 298. An ML platform team needs to prevent image overwrite. Which AWS capability is MOST appropriate?
- A) Use Immutable tags in ECR
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Immutable tags in ECR: supply chain integrity.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 299. An ML platform team needs to verify trusted inference images. Which AWS capability is MOST appropriate?
- A) Use Image signing
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Image signing: supply chain security awareness.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 300. An ML platform team needs to wait for S3 prefix in MWAA DAG. Which AWS capability is MOST appropriate?
- A) Use Airflow sensor
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Airflow sensor: orchestration dependency.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 301. An ML platform team needs to integrate external approval systems. Which AWS capability is MOST appropriate?
- A) Use SageMaker pipeline callback
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: SageMaker pipeline callback: enterprise integration.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 302. An ML platform team needs to lightweight custom logic between steps. Which AWS capability is MOST appropriate?
- A) Use Lambda step in pipeline
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Lambda step in pipeline: pipeline extensibility.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 303. An ML platform team needs to branch on metric threshold in pipeline. Which AWS capability is MOST appropriate?
- A) Use Conditional step
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Conditional step: gated promotion.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 304. An ML platform team needs to register model package after training. Which AWS capability is MOST appropriate?
- A) Use Register step
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Register step: registry integration.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 305. An ML platform team needs to create deployable model entity. Which AWS capability is MOST appropriate?
- A) Use CreateModel step
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: CreateModel step: deployment prep step.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 306. An ML platform team needs to batch score in pipeline. Which AWS capability is MOST appropriate?
- A) Use Transform step
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Transform step: validation scoring in CI.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 307. An ML platform team needs to pass evaluation report to approval step. Which AWS capability is MOST appropriate?
- A) Use PropertyFile
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: PropertyFile: metric-based gates.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 308. An ML platform team needs to Approved vs Rejected in registry. Which AWS capability is MOST appropriate?
- A) Use Model approval status
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Model approval status: governance gate.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 309. An ML platform team needs to CloudWatch alarm triggers config revert. Which AWS capability is MOST appropriate?
- A) Use Endpoint auto rollback
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Endpoint auto rollback: automated safety.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 310. An ML platform team needs to CSV, JSON, or NPY for InvokeEndpoint. Which AWS capability is MOST appropriate?
- A) Use Inference payload format
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Inference payload format: client integration detail.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 311. An ML platform team needs to required for correct deserialization. Which AWS capability is MOST appropriate?
- A) Use Content-Type header
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Content-Type header: inference API detail.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 312. An ML platform team needs to route to specific production variant. Which AWS capability is MOST appropriate?
- A) Use Target variant header
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Target variant header: testing specific variant.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 313. An ML platform team needs to enable data capture for Model Monitor. Which AWS capability is MOST appropriate?
- A) Use Capture data
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Capture data: production baseline updates.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 314. An ML platform team needs to upload labels to compare model quality. Which AWS capability is MOST appropriate?
- A) Use Ground truth labels for monitor
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Ground truth labels for monitor: supervised monitor setup.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 315. An ML platform team needs to cron on Model Monitor schedule. Which AWS capability is MOST appropriate?
- A) Use Schedule hourly monitor
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Schedule hourly monitor: continuous validation.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 316. An ML platform team needs to alert when constraints breached. Which AWS capability is MOST appropriate?
- A) Use Violations SNS
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Violations SNS: operational response.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 317. An ML platform team needs to update monitor baseline after approved drift. Which AWS capability is MOST appropriate?
- A) Use Rebaselining
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Rebaselining: lifecycle management.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 318. An ML platform team needs to cost allocation on training spend. Which AWS capability is MOST appropriate?
- A) Use Training job tags
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Training job tags: financial governance.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 319. An ML platform team needs to allocate endpoint costs. Which AWS capability is MOST appropriate?
- A) Use Inference tags
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Inference tags: financial governance.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 320. An ML platform team needs to zero invocations but running instances. Which AWS capability is MOST appropriate?
- A) Use Idle endpoint detection
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Idle endpoint detection: waste to eliminate.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 321. An ML platform team needs to RDS for feature source HA—not endpoint HA alone. Which AWS capability is MOST appropriate?
- A) Use Multi-AZ
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Multi-AZ: data source availability.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 322. An ML platform team needs to deploy across AZ with multiple instances. Which AWS capability is MOST appropriate?
- A) Use Endpoint HA
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Endpoint HA: inference availability.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 323. An ML platform team needs to InvokeEndpoint or container /ping. Which AWS capability is MOST appropriate?
- A) Use Health check
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Health check: load balancer / SageMaker health.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 324. An ML platform team needs to optimize for Inf1/GPU edge. Which AWS capability is MOST appropriate?
- A) Use Model compilation Neo
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Model compilation Neo: hardware-specific optimization.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 325. An ML platform team needs to cost-efficient inference chips. Which AWS capability is MOST appropriate?
- A) Use Inf2 instances
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Inf2 instances: instance family awareness.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 326. An ML platform team needs to GPU inference for deep learning. Which AWS capability is MOST appropriate?
- A) Use G5 instances
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: G5 instances: instance family selection.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 327. An ML platform team needs to CPU inference for classical ML. Which AWS capability is MOST appropriate?
- A) Use C6i
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: C6i: right-size compute family.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 328. An ML platform team needs to large embedding models. Which AWS capability is MOST appropriate?
- A) Use Memory-optimized
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Memory-optimized: instance family for memory-bound.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 329. An ML platform team needs to large payload async/batch needs. Which AWS capability is MOST appropriate?
- A) Use Network bandwidth
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Network bandwidth: infrastructure sizing factor.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 330. An ML platform team needs to VPC endpoint scaling constraint. Which AWS capability is MOST appropriate?
- A) Use ENI limits
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: ENI limits: VPC planning awareness.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 331. An ML platform team needs to plan CIDR for scaled endpoints. Which AWS capability is MOST appropriate?
- A) Use Subnet IP exhaustion
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Subnet IP exhaustion: VPC capacity planning.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 332. An ML platform team needs to minimize via VPC endpoints. Which AWS capability is MOST appropriate?
- A) Use NAT cost
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: NAT cost: cost-aware networking.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 333. An ML platform team needs to centralized networking for SageMaker. Which AWS capability is MOST appropriate?
- A) Use Shared VPC
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Shared VPC: enterprise network pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
Question 334. An ML platform team needs to central ML platform account. Which AWS capability is MOST appropriate?
- A) Use Hub account deployment
- B) Use AWS DeepRacer exclusively
- C) Disable IAM and encryption
- D) Store artifacts only on local laptops
Correct answer: A
Why this is correct: Hub account deployment: multi-account ML pattern.
Why the distractors are incorrect:
- B) Use AWS DeepRacer exclusively — Out of exam scope or violates security and governance requirements.
- C) Disable IAM and encryption — Out of exam scope or violates security and governance requirements.
- D) Store artifacts only on local laptops — Out of exam scope or violates security and governance requirements.
How to use this bank
- Study by domain weights (28 / 26 / 22 / 24).
- For each wrong answer, write why the scenario rejects it—exam distractors are plausible.
- Lab: SageMaker Pipelines, Model Registry, endpoint types, Model Monitor, CodePipeline integration.
Related posts
Machine Learning Foundations · Generative AI Foundations · Data Engineering · AI and ML terminology