Google Professional Machine Learning Engineer Practice Test 89
Question 880
During a postdeployment AI safety and reliability review, the following condition occurs: A model keeps similar average scores after an upstream data migration, but feature importance shifts from stable signals toward a suspicious proxy variable. The model owner must provide repeatable evidence rather than a cosmetic dashboard change. Which statement identifies the primary reason for this failure?
- Normal recurring seasonality can resemble drift against an unsegmented static baseline.
- Prompt wording alone is not a reliable comprehensive safety boundary for public generative applications.
- Feature attribution drift can indicate changing model reliance even when overall score distributions remain stable.
- Predictive importance is not causal proof; attribution techniques describe model behavior under their assumptions.
Explanation
Correct Answer
C:Feature attribution drift can indicate changing model reliance even when overall score distributions remain stable. This explains the mechanism behind feature attribution drift after deployment and why the described failure is possible. A discriminating investigation is: Examine explanation distributions, feature generation changes, protected attribute proxies and prediction outcomes by cohort. The corresponding repair is: Investigate attribution drift, remove invalid proxy features if warranted and retrain with appropriate validation.
Incorrect Answers
A:Normal recurring seasonality can resemble drift against an unsegmented static baseline. This principle describes model monitoring false-alert seasonality, which is not the cause of the observed feature attribution drift after deployment failure. The controlling explanation is: Feature attribution drift can indicate changing model reliance even when overall score distributions remain stable.
B:Prompt wording alone is not a reliable comprehensive safety boundary for public generative applications. This principle describes gen ai output safety policy enforcement, which is not the cause of the observed feature attribution drift after deployment failure. The controlling explanation is: Feature attribution drift can indicate changing model reliance even when overall score distributions remain stable.
D:Predictive importance is not causal proof; attribution techniques describe model behavior under their assumptions. This principle describes explainability versus causal claim, which is not the cause of the observed feature attribution drift after deployment failure. The controlling explanation is: Feature attribution drift can indicate changing model reliance even when overall score distributions remain stable.
Question 881
During a postdeployment AI safety and reliability review, the following condition occurs: A product support bot submits raw customer payment and identity records to a generative service even though the question only needs an order status summary. The model owner must provide repeatable evidence rather than a cosmetic dashboard change. Which underlying engineering principle explains why the original approach fails?
- Normal recurring seasonality can resemble drift against an unsegmented static baseline.
- Sensitive information should be shared only when necessary and permitted; ordinary prompting is not an exemption from privacy controls.
- Predictive importance is not causal proof; attribution techniques describe model behavior under their assumptions.
- External retrieval content can carry attacker instructions; data should not receive the authority of trusted application instructions.
Explanation
Correct Answer
B:Sensitive information should be shared only when necessary and permitted; ordinary prompting is not an exemption from privacy controls. This explains the mechanism behind sensitive identifiers leaked to llm prompts and why the described failure is possible. A discriminating investigation is: Inspect input logs, sensitivity classifications, prompt assembly and approved data-processing boundaries. The corresponding repair is: Minimize or tokenize personal fields before generation and enforce access and retention controls on any necessary context.
Incorrect Answers
A:Normal recurring seasonality can resemble drift against an unsegmented static baseline. This principle describes model monitoring false-alert seasonality, which is not the cause of the observed sensitive identifiers leaked to llm prompts failure. The controlling explanation is: Sensitive information should be shared only when necessary and permitted; ordinary prompting is not an exemption from privacy controls.
C:Predictive importance is not causal proof; attribution techniques describe model behavior under their assumptions. This principle describes explainability versus causal claim, which is not the cause of the observed sensitive identifiers leaked to llm prompts failure. The controlling explanation is: Sensitive information should be shared only when necessary and permitted; ordinary prompting is not an exemption from privacy controls.
D:External retrieval content can carry attacker instructions; data should not receive the authority of trusted application instructions. This principle describes prompt injection through retrieved documents, which is not the cause of the observed sensitive identifiers leaked to llm prompts failure. The controlling explanation is: Sensitive information should be shared only when necessary and permitted; ordinary prompting is not an exemption from privacy controls.
Question 882
As the team tests a new governed AI release against adversarial and normal data, the following condition occurs: A production evaluator cannot match delayed loan repayment outcomes to the predictions that were made weeks earlier because requests lack stable IDs. The fix must retain legitimate model usefulness and appropriate controls. Which outcome most directly demonstrates a successful, measurable fix?
- An evaluation fixture correctly associates each matured label with exactly one prior prediction.
- A failing canary halts and restores the healthy model before broader traffic exposure.
- The authorized pipeline still reads its intended data while attempts to export to unapproved destinations are denied.
- A synthetic distribution shift triggers an attributable alert while an unchanged control sample stays within tolerance.
Explanation
Correct Answer
A:An evaluation fixture correctly associates each matured label with exactly one prior prediction. This is specific acceptance evidence for ground truth association for quality monitoring. Supervised production quality requires reliable prediction-to-ground-truth linkage, not only unlabeled model outputs. It confirms the controlled correction: Capture immutable prediction identifiers and join verified outcome records to the corresponding earlier inference request. This matters here because as the team tests a new governed AI release against adversarial and normal data, the following condition occurs: A….
Incorrect Answers
B:A failing canary halts and restores the healthy model before broader traffic exposure. This result could demonstrate success for production incident rollback based on metrics but not for the stated ground truth association for quality monitoring issue. The direct test is: An evaluation fixture correctly associates each matured label with exactly one prior prediction. This matters here because as the team tests a new governed AI release against adversarial and normal data, the following condition occurs: A….
C:The authorized pipeline still reads its intended data while attempts to export to unapproved destinations are denied. This result could demonstrate success for training data exfiltration through broad access but not for the stated ground truth association for quality monitoring issue. The direct test is: An evaluation fixture correctly associates each matured label with exactly one prior prediction.
D:A synthetic distribution shift triggers an attributable alert while an unchanged control sample stays within tolerance. This result could demonstrate success for input data distribution drift alarm but not for the stated ground truth association for quality monitoring issue. The direct test is: An evaluation fixture correctly associates each matured label with exactly one prior prediction.
Question 883
During model-training validation on a Google Cloud experimental project, the following condition occurs: A hospital alert model achieves 99% accuracy by predicting every patient is negative when the positive condition occurs in 1% of cases. The change should preserve a meaningful independent model-quality comparison. Which targeted engineering change most directly addresses the failure?
- Compare an interpretable model or validated feature attribution approach with the black box and document decision tradeoffs.
- Split raw records first and resample only inside training folds, leaving validation cohorts untouched.
- Choose a decision threshold using acceptable recall and alert burden with a validated independent cohort.
- Use grounded retrieval for frequently changing policy facts and reserve fine-tuning for measured durable behavior adaptation.
Explanation
Correct Answer
C:Choose a decision threshold using acceptable recall and alert burden with a validated independent cohort. This is the targeted corrective response for cost-sensitive threshold for rare positives. Accuracy on highly imbalanced labels can hide severe minority-class recall failures. Verify the diagnosis using: Inspect prevalence, confusion matrix, false-negative harm, calibration and precision-recall curves at candidate thresholds. Confirm success using: The chosen threshold detects clinically relevant positives while the false-alarm rate is documented for deployment review. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A hospital….
Incorrect Answers
A:Compare an interpretable model or validated feature attribution approach with the black box and document decision tradeoffs. This action addresses interpretability-constrained model baseline, not the failure in the described cost-sensitive threshold for rare positives case. The targeted correction is: Choose a decision threshold using acceptable recall and alert burden with a validated independent cohort. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A hospital….
B:Split raw records first and resample only inside training folds, leaving validation cohorts untouched. This action addresses training class imbalance fold leakage, not the failure in the described cost-sensitive threshold for rare positives case. The targeted correction is: Choose a decision threshold using acceptable recall and alert burden with a validated independent cohort.
D:Use grounded retrieval for frequently changing policy facts and reserve fine-tuning for measured durable behavior adaptation. This action addresses foundation model prompt versus fine-tuning, not the failure in the described cost-sensitive threshold for rare positives case. The targeted correction is: Choose a decision threshold using acceptable recall and alert burden with a validated independent cohort. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A hospital….
Question 884
Within a jointly maintained Google Cloud data science workspace, the following condition occurs: A team imports a pretrained JAX checkpoint in a notebook but assumes its model weights and device setup will load unchanged in a PyTorch-only runtime. Collaborators must avoid expanding data access or breaking experiment reproducibility. What acceptance result verifies the intended behavior after remediation?
- Annotators can complete validated labels without seeing identifying data and the project retains compliant traceability.
- Independent notebook runs resolve the same approved libraries and produce comparable metric and artifact outputs.
- A clean notebook loads the exact model revision, runs representative inputs and reports consistent outputs.
- The exploration completes within notebook memory while its aggregate statistics match trusted BigQuery queries.
Explanation
Correct Answer
C:A clean notebook loads the exact model revision, runs representative inputs and reports consistent outputs. This is specific acceptance evidence for model garden jax accelerator compatibility. Framework-native checkpoints and runtime APIs are not automatically interchangeable across ML libraries. It confirms the controlled correction: Choose a compatible Model Garden model/runtime or perform a validated checkpoint conversion before training. This matters here because within a jointly maintained Google Cloud data science workspace, the following condition occurs: A team imports a….
Incorrect Answers
A:Annotators can complete validated labels without seeing identifying data and the project retains compliant traceability. This result could demonstrate success for data minimization for labeling inputs but not for the stated model garden jax accelerator compatibility issue. The direct test is: A clean notebook loads the exact model revision, runs representative inputs and reports consistent outputs.
B:Independent notebook runs resolve the same approved libraries and produce comparable metric and artifact outputs. This result could demonstrate success for notebook dependency lock reproducibility but not for the stated model garden jax accelerator compatibility issue. The direct test is: A clean notebook loads the exact model revision, runs representative inputs and reports consistent outputs.
D:The exploration completes within notebook memory while its aggregate statistics match trusted BigQuery queries. This result could demonstrate success for notebook memory exhaustion on full data extraction but not for the stated model garden jax accelerator compatibility issue. The direct test is: A clean notebook loads the exact model revision, runs representative inputs and reports consistent outputs. This matters here because within a jointly maintained Google Cloud data science workspace, the following condition occurs: A team imports a….
Question 885
During a production ML inference release on Google Cloud, the following condition occurs: A private inference service exists with proper invoker IAM, but a VPC client cannot connect because the private route and DNS mapping are missing. The service must meet its agreed quality and latency requirements. What implementation change resolves the specific issue without substituting an unrelated fix?
- Configure the supported private DNS, network route and endpoint access path without granting public ingress.
- Enable compatible dynamic batching with a bounded wait time and benchmark throughput against p95 latency.
- Preserve stable canonical identifier types end-to-end and backfill only the affected feature records.
- Roll out a small canary percentage first, monitor error and quality guardrails and expand only after acceptance.
Explanation
Correct Answer
A:Configure the supported private DNS, network route and endpoint access path without granting public ingress. This is the targeted corrective response for private vpc routing failure for serving. Identity permission is not proof of network reachability; private serving also requires correct routing and name resolution. Verify the diagnosis using: Inspect resolved endpoint address, authorized network routing, service access policy and connection logs. Confirm success using: A test client on the approved network succeeds while an external probe remains blocked. This matters here because a private inference service exists with proper invoker IAM, but a VPC client cannot connect because the private….
Incorrect Answers
B:Enable compatible dynamic batching with a bounded wait time and benchmark throughput against p95 latency. This action addresses dynamic batch serving for gpu saturation, not the failure in the described private vpc routing failure for serving case. The targeted correction is: Configure the supported private DNS, network route and endpoint access path without granting public ingress.
C:Preserve stable canonical identifier types end-to-end and backfill only the affected feature records. This action addresses feature online entity key mismatch, not the failure in the described private vpc routing failure for serving case. The targeted correction is: Configure the supported private DNS, network route and endpoint access path without granting public ingress.
D:Roll out a small canary percentage first, monitor error and quality guardrails and expand only after acceptance. This action addresses canary gradual traffic rollout, not the failure in the described private vpc routing failure for serving case. The targeted correction is: Configure the supported private DNS, network route and endpoint access path without granting public ingress.
Question 886
During model-training validation on a Google Cloud experimental project, the following condition occurs: A weeklong custom training job uses preemptible capacity but saves checkpoints only to a temporary local container filesystem. The change should preserve a meaningful independent model-quality comparison. What acceptance result verifies the intended behavior after remediation?
- New warranty rules appear correctly after the index update without unnecessary retraining and sources remain traceable.
- The training manifest contains traceable permitted sources and safety tests detect known prohibited outputs.
- A controlled interruption resumes near the saved step without restarting training from zero.
- The deployed artifact corresponds to the best held-out epoch rather than the final overfit epoch.
Explanation
Correct Answer
C:A controlled interruption resumes near the saved step without restarting training from zero. This is specific acceptance evidence for training checkpoint persistence across interruptions. Preemptible execution can end abruptly; local-only state is not a durable recovery mechanism. It confirms the controlled correction: Persist versioned checkpoints in durable Cloud Storage and make the training code resume optimizer and model state. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A weeklong….
Incorrect Answers
A:New warranty rules appear correctly after the index update without unnecessary retraining and sources remain traceable. This result could demonstrate success for foundation model prompt versus fine-tuning but not for the stated training checkpoint persistence across interruptions issue. The direct test is: A controlled interruption resumes near the saved step without restarting training from zero.
B:The training manifest contains traceable permitted sources and safety tests detect known prohibited outputs. This result could demonstrate success for fine-tuning image generator data rights but not for the stated training checkpoint persistence across interruptions issue. The direct test is: A controlled interruption resumes near the saved step without restarting training from zero. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A weeklong….
D:The deployed artifact corresponds to the best held-out epoch rather than the final overfit epoch. This result could demonstrate success for early stopping on validation divergence but not for the stated training checkpoint persistence across interruptions issue. The direct test is: A controlled interruption resumes near the saved step without restarting training from zero.
Question 887
While reproducing a training orchestration incident in isolated staging, the following condition occurs: A production workflow starts model registration before validation and evaluation steps have finished, occasionally promoting untested weights. The team requires an idempotent pipeline behavior and traceable evidence. Select TWO: the decisive diagnostic step and the targeted repair for this specific issue.
- Inspect ML Metadata execution links, table snapshots, transformations and model version lineage.
- Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs.
- Train only on adequately matured labeled cohorts and monitor recent feature drift separately until outcomes arrive.
- Add explicit upstream validation and evaluation dependencies before registration and deployment components.
- Package the fitted preprocessing component with the model and apply the same validated transform in both paths.
Explanation
Correct Answers
B:Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs. This inspects evidence relevant to pipeline component dependency order. Orchestration needs explicit data and control dependencies; notebook execution order is not a production DAG. After identifying the cause, the matching correction is: Add explicit upstream validation and evaluation dependencies before registration and deployment components. This matters here because while reproducing a training orchestration incident in isolated staging, the following condition occurs: A….
D:Add explicit upstream validation and evaluation dependencies before registration and deployment components. This is the targeted corrective response for pipeline component dependency order. Orchestration needs explicit data and control dependencies; notebook execution order is not a production DAG. Verify the diagnosis using: Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs. Confirm success using: Every production model version traces to successfully completed validation and evaluation steps in that run. This matters here because while reproducing a training orchestration incident in isolated staging, the following condition occurs: A….
Incorrect Answers
A:Inspect ML Metadata execution links, table snapshots, transformations and model version lineage. This is an action or diagnostic relevant to model feature lineage for audit, not one of the required two steps. In this scenario verify: Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs. Then correct: Add explicit upstream validation and evaluation dependencies before registration and deployment components.
C:Train only on adequately matured labeled cohorts and monitor recent feature drift separately until outcomes arrive. This is an action or diagnostic relevant to scheduled retraining on delayed ground truth, not one of the required two steps. In this scenario verify: Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs. Then correct: Add explicit upstream validation and evaluation dependencies before registration and deployment components.
E:Package the fitted preprocessing component with the model and apply the same validated transform in both paths. This is an action or diagnostic relevant to training serving preprocessing artifact sharing, not one of the required two steps. In this scenario verify: Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs. Then correct: Add explicit upstream validation and evaluation dependencies before registration and deployment components. This matters here because while reproducing a training orchestration incident in isolated staging, the following condition occurs: A….
Question 888
At a Google Cloud analytics team preparing a governed pilot, the following condition occurs: An accounts-payable team needs supplier names, invoice totals and line items but uses image classification to decide whether an invoice is present. The team must satisfy a reproducible low-code AI assessment. Which outcome most directly demonstrates a successful, measurable fix?
- A held-out invoice batch produces correct structured fields with measurable per-field precision and reject handling.
- A revised policy becomes answerable after document indexing without retraining, and retrieved citations match the active version.
- On held-out questions, answers remain grounded while input tokens and p95 response latency decline.
- Predictions contain sensible numeric estimates and holdout MAE or RMSE is measured against actual monetary values.
Explanation
Correct Answer
A:A held-out invoice batch produces correct structured fields with measurable per-field precision and reject handling. This is specific acceptance evidence for document ai structured invoice extraction. Object or image classification only identifies image categories; document understanding targets structured text fields. It confirms the controlled correction: Use an appropriate Document AI processor for form or invoice extraction and validate required fields before downstream posting.
Incorrect Answers
B:A revised policy becomes answerable after document indexing without retraining, and retrieved citations match the active version. This result could demonstrate success for foundation model tuning versus retrieval update but not for the stated document ai structured invoice extraction issue. The direct test is: A held-out invoice batch produces correct structured fields with measurable per-field precision and reject handling.
C:On held-out questions, answers remain grounded while input tokens and p95 response latency decline. This result could demonstrate success for gemini prompt context token budget but not for the stated document ai structured invoice extraction issue. The direct test is: A held-out invoice batch produces correct structured fields with measurable per-field precision and reject handling.
D:Predictions contain sensible numeric estimates and holdout MAE or RMSE is measured against actual monetary values. This result could demonstrate success for bigquery ml linear regression for continuous value but not for the stated document ai structured invoice extraction issue. The direct test is: A held-out invoice batch produces correct structured fields with measurable per-field precision and reject handling.
Question 889
While reproducing a training orchestration incident in isolated staging, the following condition occurs: An automated retraining job overwrites the previous serving model before verifying quality under the most recent data distribution. The team requires an idempotent pipeline behavior and traceable evidence. What acceptance result verifies the intended behavior after remediation?
- Training manifests exclude unlabeled recent transactions and measured metrics use verified ground truth.
- An intentionally failing candidate is rolled back promptly and the previous version remains serving.
- Running the same task twice produces one logical batch and preserves prior validated partitions.
- No-op periods create no unnecessary models and validated drift events produce candidates requiring approval.
Explanation
Correct Answer
B:An intentionally failing candidate is rolled back promptly and the previous version remains serving. This is specific acceptance evidence for continuous training rollback protection. Continuous training should not imply uncontrolled continuous deployment of every new model. It confirms the controlled correction: Retain the last healthy model and require canary validation before switching all prediction traffic. This matters here because while reproducing a training orchestration incident in isolated staging, the following condition occurs: An….
Incorrect Answers
A:Training manifests exclude unlabeled recent transactions and measured metrics use verified ground truth. This result could demonstrate success for scheduled retraining on delayed ground truth but not for the stated continuous training rollback protection issue. The direct test is: An intentionally failing candidate is rolled back promptly and the previous version remains serving. This matters here because while reproducing a training orchestration incident in isolated staging, the following condition occurs: An….
C:Running the same task twice produces one logical batch and preserves prior validated partitions. This result could demonstrate success for airflow task idempotency on retry but not for the stated continuous training rollback protection issue. The direct test is: An intentionally failing candidate is rolled back promptly and the previous version remains serving.
D:No-op periods create no unnecessary models and validated drift events produce candidates requiring approval. This result could demonstrate success for data drift triggered retraining policy but not for the stated continuous training rollback protection issue. The direct test is: An intentionally failing candidate is rolled back promptly and the previous version remains serving. This matters here because while reproducing a training orchestration incident in isolated staging, the following condition occurs: An….
Question 890
During a production ML inference release on Google Cloud, the following condition occurs: A multi-class model returns numeric class indices, but an API gateway uses a stale alphabetical mapping and mislabels predictions. The service must meet its agreed quality and latency requirements. What acceptance result verifies the intended behavior after remediation?
- The running endpoint resolves to a traceable evaluated package and cannot silently switch weights when an alias moves.
- Every controlled sample's returned label matches the model's trained class index with no misordered classes.
- A controlled failed rollout restores the prior model and returns errors below the service objective.
- Previously missed customer IDs retrieve the intended feature values without collapsing distinct leading-zero identifiers.
Explanation
Correct Answer
B:Every controlled sample's returned label matches the model's trained class index with no misordered classes. This is specific acceptance evidence for custom postprocessing label mapping drift. Index ordering is part of the model contract; separate stale label dictionaries can silently corrupt inference. It confirms the controlled correction: Ship the class vocabulary with the model artifact and version or validate the output adapter against it. This matters here because a multi-class model returns numeric class indices, but an API gateway uses a stale alphabetical mapping and….
Incorrect Answers
A:The running endpoint resolves to a traceable evaluated package and cannot silently switch weights when an alias moves. This result could demonstrate success for model registry version provenance but not for the stated custom postprocessing label mapping drift issue. The direct test is: Every controlled sample's returned label matches the model's trained class index with no misordered classes.
C:A controlled failed rollout restores the prior model and returns errors below the service objective. This result could demonstrate success for model traffic rollback on high error rate but not for the stated custom postprocessing label mapping drift issue. The direct test is: Every controlled sample's returned label matches the model's trained class index with no misordered classes.
D:Previously missed customer IDs retrieve the intended feature values without collapsing distinct leading-zero identifiers. This result could demonstrate success for feature online entity key mismatch but not for the stated custom postprocessing label mapping drift issue. The direct test is: Every controlled sample's returned label matches the model's trained class index with no misordered classes. This matters here because a multi-class model returns numeric class indices, but an API gateway uses a stale alphabetical mapping and….