Architecting low-code AI solutions
Free Google Cloud Professional Machine Learning Engineer practice — 6 questions on Architecting low-code AI solutions, with explanations. No sign-up. Full 12-question mixed test →
A retail company's transaction data already lives in BigQuery. The data science team wants to train a churn-prediction model using AutoML's automated feature engineering and neural architecture search, but the ML engineer insists on keeping the entire workflow in SQL without exporting data or moving to the Vertex AI console. Which approach satisfies this requirement?
BigQuery ML supports training AutoML Tables models directly from SQL using model_type='AUTOML_CLASSIFIER' or 'AUTOML_REGRESSOR' in CREATE MODEL, giving automated feature engineering and architecture search without leaving BigQuery or exporting data.
A finance team receives invoices from thousands of different vendors, each with a distinct layout. They need to extract structured fields like invoice number, vendor name, line items, and total amount with high accuracy and minimal engineering effort. Which solution BEST meets this requirement?
Document AI provides specialized, pretrained processors for common document types including invoices, which already understand key-value fields like invoice number, vendor, and totals across varying layouts — requiring no training data or custom labeling.
A team trains a BigQuery ML linear regression model using CREATE MODEL with a TRANSFORM clause that includes ML.STANDARD_SCALER and ML.BUCKETIZE on input columns. At prediction time, they call ML.PREDICT on new raw data. What happens to the preprocessing logic defined in TRANSFORM?
The TRANSFORM clause in BigQuery ML CREATE MODEL captures preprocessing steps and persists them with the model. ML.PREDICT automatically applies these same transformations to new input data, ensuring consistency between training and serving and avoiding train/serve skew.
An ML engineer is evaluating Model Garden in Vertex AI as part of a low-code strategy. Which statement BEST describes what Model Garden provides compared to AutoML?
Model Garden is Vertex AI's model catalog offering access to Google first-party foundation models, third-party partner models, and open-source models — enabling discovery, testing, one-click deployment, or fine-tuning, in contrast to AutoML which trains a custom model from your own labeled dataset.
A video-sharing platform needs to automatically flag uploaded videos containing explicit or violent content at scale, with results needed within hours of launch and no labeled training data available. Which approach BEST meets this requirement?
The Video Intelligence API includes a pretrained Explicit Content Detection feature specifically designed to flag adult/violent content in videos out of the box, requiring no training data and enabling immediate deployment.
A developer is preparing a minimal proof-of-concept using AutoML Vision (single-label image classification) and needs to know the exam-tested minimum data requirement. What is the minimum number of training images required per label to successfully train an AutoML Vision image classification model (though Google recommends far more for production accuracy)?
AutoML Vision requires a minimum of 10 training images per label to begin training a single-label image classification model, though Google explicitly recommends at least 100 images per label for reliable production-quality accuracy.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$129.99 $34.99 with code FREETEST33 — valid through September 7.
Get my $34.99 deal →