Google Professional Machine Learning Engineer PMLE practice test

A travel booking company has a real-time recommendation system that suggests hotels and flights to users. The model is served using TensorFlow Serving on a Google Kubernetes Engine (GKE) cluster with auto-scaling enabled. The cluster uses n1-standard-4 machine types. The team has set up Cloud Monitoring dashboards and alerts. Last week, during a major holiday promotion, the team noticed that the model's inference latency P99 increased from 150 ms to 450 ms over a 30-minute period, while the request throughput increased from 500 to 1,200 requests per second. CPU utilization across the cluster rose to 95%, but memory utilization remained at 60%. The model version and the serving infrastructure configuration have not changed since the last deployment. Which action should the team take to mitigate the latency issue?

Trap 1: Implement a feature engineering pipeline that compresses the input…

While potentially beneficial, this is a longer-term solution and does not provide immediate latency relief during the surge.

Trap 2: Deploy a newer version of the model that uses a more efficient…

Deploying a new model requires time for development, testing, and approval, and may not be feasible for immediate mitigation.

Trap 3: Increase the number of TensorFlow Serving instances by reducing the…

Reducing CPU requests may lead to CPU starvation and pod instability, harming latency further.

A
Implement a feature engineering pipeline that compresses the input features to reduce data size and inference time.
Why wrong: While potentially beneficial, this is a longer-term solution and does not provide immediate latency relief during the surge.
B
Deploy a newer version of the model that uses a more efficient architecture to reduce computational complexity.
Why wrong: Deploying a new model requires time for development, testing, and approval, and may not be feasible for immediate mitigation.
C
Increase the number of TensorFlow Serving instances by reducing the CPU request per pod in GKE to allow more pods per node.
Why wrong: Reducing CPU requests may lead to CPU starvation and pod instability, harming latency further.
D
Add more nodes to the GKE cluster to increase the total CPU resources available for serving.
Adding nodes increases compute capacity, allowing more parallel inference and reducing latency under high load.

Question 2easymultiple choice

A global retail company uses Vertex AI Recommendations to provide product recommendations on their website. They have a large catalog and millions of users. The initial deployment works well for active users, but they notice that new users (with no purchase history) receive generic recommendations that are not personalized. The company wants to improve the cold-start experience. They have user demographic data (age, location) available at sign-up. Current recommendation model is a collaborative filtering model using the built-in Vertex AI Recommendations. What should the company do to improve personalization for new users?

Trap 1: Collect more historical interaction data before showing…

New users have no history; waiting does not help.

Trap 2: Disable recommendations for new users until they have at least 10…

This would lose the opportunity to engage new users.

Trap 3: Build a custom two-tower recommendation model using Vertex AI…

Building a custom model is more complex and may not be needed.

A
Collect more historical interaction data before showing recommendations
Why wrong: New users have no history; waiting does not help.
B
Disable recommendations for new users until they have at least 10 interactions
Why wrong: This would lose the opportunity to engage new users.
C
Increase the user exploration parameter in the Vertex AI Recommendations configuration
Exploration helps serve diverse items to new users to learn preferences.
D
Build a custom two-tower recommendation model using Vertex AI Training
Why wrong: Building a custom model is more complex and may not be needed.

Question 3mediummultiple choice

Your team is developing a machine learning model for real-time fraud detection. The training pipeline runs on Vertex AI and uses BigQuery for feature engineering. Recently, the pipeline has been taking significantly longer to execute. Upon investigation, you find that the BigQuery query for feature extraction is being rerun every time the pipeline runs, even though the underlying data hasn't changed. The pipeline is scheduled to run every hour. You want to reduce cost and execution time without losing the ability to detect data drifts. Which approach should you take?

Trap 1: Implement a caching mechanism in the pipeline that stores the…

Pipeline caching is based on component inputs, not on data content, so it may not prevent rerun if inputs differ.

Trap 2: Reduce the pipeline frequency to once a day to minimize the number…

This reduces cost but delays model updates and data drift detection.

Trap 3: Use a conditional pipeline that checks if the data has changed…

This adds complexity and still requires executing the pipeline to perform the check.

A
Implement a caching mechanism in the pipeline that stores the results of the BigQuery query and reuses them if the data hasn't changed.
Why wrong: Pipeline caching is based on component inputs, not on data content, so it may not prevent rerun if inputs differ.
B
Move the feature extraction to a separate scheduled query in BigQuery and load the results into a table that the pipeline reads from.
This separates concerns and avoids redundant execution, while still allowing data drift detection via the pipeline.
C
Reduce the pipeline frequency to once a day to minimize the number of runs.
Why wrong: This reduces cost but delays model updates and data drift detection.
D
Use a conditional pipeline that checks if the data has changed before running the feature extraction step.
Why wrong: This adds complexity and still requires executing the pipeline to perform the check.

Read the full NAT/PAT explanation →

Question 4mediummultiple choice

A healthcare organization is building a machine learning model to predict patient readmission risk. They have sensitive data stored in BigQuery that includes protected health information (PHI). The data science team uses Vertex AI Workbench notebooks to explore the data and develop models. The organization's security policy requires that all PHI data must be encrypted at rest and in transit, and that access to the data is logged and audited. They also need to ensure that the data used for model training is de-identified to remove direct identifiers such as patient names and SSNs. The team wants to automate the de-identification process as part of the data pipeline. Which approach meets these requirements?

Trap 1: Enable Shielded VM on Vertex AI Workbench notebooks and use VPC-SC…

Shielded VM and VPC-SC provide security but do not de-identify data.

Trap 2: Use Cloud Key Management Service to encrypt the PHI columns in…

Encryption does not remove identifiers; the team would still see PHI after decryption.

Trap 3: Use BigQuery row-level security to mask PHI columns for the data…

Row-level security does not remove identifiers for training; it only masks at query time.

A
Create a Dataflow pipeline that reads from the original BigQuery table, applies Cloud DLP de-identification transforms, and writes to a new BigQuery table. Grant the data science team access to the de-identified table.
Dataflow with DLP automates de-identification and creates a safe dataset.
B
Enable Shielded VM on Vertex AI Workbench notebooks and use VPC-SC to restrict data access.
Why wrong: Shielded VM and VPC-SC provide security but do not de-identify data.
C
Use Cloud Key Management Service to encrypt the PHI columns in BigQuery, and share the encryption key with the data science team.
Why wrong: Encryption does not remove identifiers; the team would still see PHI after decryption.
D
Use BigQuery row-level security to mask PHI columns for the data science team, and train the model directly on the original table.
Why wrong: Row-level security does not remove identifiers for training; it only masks at query time.

Question 5hardmultiple choice

You are an ML engineer at a global e-commerce company. Your team has developed a deep learning model for product recommendation that runs on Vertex AI Prediction. The model is deployed on a single n1-highmem-2 instance (CPU only) with autoscaling enabled (min replicas=1, max replicas=10). During Black Friday, traffic spikes to 1000 requests per second (QPS), and you observe that latency increases from 50ms to over 5000ms, and many requests time out. You check the monitoring dashboard and see that CPU utilization is at 100% on the single instance, and autoscaling is not triggering quickly enough. The team has a budget for this service and wants to handle the spike without compromising latency. What should you do?

Trap 1: Increase min replicas to 5 to keep warm instances

Without improving per-instance throughput, warm instances may still be insufficient.

Trap 2: Set min replicas=1 and max replicas=5 to control cost

Limiting max replicas may not handle the spike.

Trap 3: Increase max replicas to 20 and keep CPU instances

CPU instances have high latency per request; more replicas may not reduce latency enough.

A
Switch to GPU instances (e.g., n1-standard-4 with T4) and set min replicas=2 with autoscaling up to 10
GPUs accelerate inference, reducing per-request latency; warm instances handle spike.
B
Increase min replicas to 5 to keep warm instances
Why wrong: Without improving per-instance throughput, warm instances may still be insufficient.
C
Set min replicas=1 and max replicas=5 to control cost
Why wrong: Limiting max replicas may not handle the spike.
D
Increase max replicas to 20 and keep CPU instances
Why wrong: CPU instances have high latency per request; more replicas may not reduce latency enough.

Question 6mediummultiple choice

A financial services company uses Vertex AI AutoML Tables to build a credit risk model. The dataset contains 500,000 rows and 50 features, including loan amount, credit score, debt-to-income ratio, and employment length. The target variable is binary: 'default' (1) or 'no default' (0). The data is highly imbalanced, with only 2% defaults. The data scientist trains a model with AutoML Tables using default settings. The evaluation metrics show an AUC of 0.85, but the confusion matrix reveals that the model predicts 'no default' for almost all cases, missing most defaults. The data scientist needs to improve the model's ability to identify defaults without significantly increasing false positives. They have limited time and cannot write custom code. What should they do?

Trap 1: Manually split the data into a stratified train/test set to ensure…

Why B is wrong: AutoML already does stratified splitting; this won't improve recall.

Trap 2: Train multiple models with different algorithms (e.g., XGBoost,…

Why C is wrong: This requires custom code, not low-code.

Trap 3: Under-sample the majority class to create a balanced dataset and…

Why D is wrong: Under-sampling may lose valuable data and is not recommended with AutoML.

A
Manually split the data into a stratified train/test set to ensure the same proportion of defaults in each.
Why wrong: Why B is wrong: AutoML already does stratified splitting; this won't improve recall.
B
Train multiple models with different algorithms (e.g., XGBoost, Random Forest) and blend them using a custom script.
Why wrong: Why C is wrong: This requires custom code, not low-code.
C
Enable 'Enable weighted evaluation' and set the optimization objective to 'Maximize recall at a specific recall@P%' with a target precision of 0.5.
Why A is correct: AutoML Tables supports custom optimization objectives to handle imbalance.
D
Under-sample the majority class to create a balanced dataset and retrain.
Why wrong: Why D is wrong: Under-sampling may lose valuable data and is not recommended with AutoML.