GOOGLE CLOUD · PMLE

Google Cloud Professional Machine Learning Engineer practice questions

ML engineers who build and run models on Google Cloud

Pass markPass or fail (score not published)
Questions in the exam50–60
Time allowed120 min
Exam fee$200
Practice questions here25

What the PMLE exam covers

Current exam guide
Architecting low-code AI solutions13%
Collaborating to manage data and models16%
Scaling prototypes into ML models21%
Serving and scaling models20%
Automating and orchestrating ML pipelines18%
Monitoring AI solutions13%

Sample PMLE questions

Five of the 25 practice questions, each with its answer and explanation.

Architecting low-code AI solutions

Analysts with data already in BigQuery want to build a regression model using SQL. Which option fits?

  1. Cloud DNS
  2. Apigee
  3. BigQuery ML
  4. A custom GPU cluster
Show the answer

C. BigQuery ML

BigQuery ML trains models with SQL where the data already lives.

Architecting low-code AI solutions

A team needs text and fields extracted from scanned forms with minimal ML work. Which option fits?

  1. Training a model from scratch
  2. Cloud CDN
  3. Document AI
  4. BigQuery scheduled queries
Show the answer

C. Document AI

Pre-trained APIs such as Document AI solve common tasks without custom training.

Architecting low-code AI solutions

A business team wants a tabular classification model without writing model code. Which option fits?

  1. AutoML on Google's AI platform
  2. Cloud Armor
  3. Pub/Sub
  4. Writing a custom TensorFlow model
Show the answer

A. AutoML on Google's AI platform

AutoML trains and tunes models from tabular data with little code.

Collaborating to manage data and models

Large volumes of data need preprocessing in both batch and streaming modes. Which service fits?

  1. Cloud Billing
  2. Dataflow
  3. Cloud Scheduler alone
  4. Cloud Functions only
Show the answer

B. Dataflow

Dataflow runs scalable batch and streaming pipelines.

Collaborating to manage data and models

Training data contains personal information that must not reach the model. What should happen?

  1. De-identify it before training, for example with Sensitive Data Protection
  2. Train on it and delete the model later
  3. Rename the columns
  4. Encrypt only the model
Show the answer

A. De-identify it before training, for example with Sensitive Data Protection

Removing or masking PII before training prevents it being learned or exposed.