GCP ML Engineer (Google Cloud Certifications) flashcards
GCP ML Engineer · Build recall with a focused review session.
32 cards. Read the term and the definition here. The deck at the top of this page is optional.
- Vertex AI AutoML
- A managed Vertex AI service that trains image, text, tabular, and video models from labeled data without requiring custom training code.
- BigQuery ML
- A BigQuery capability that lets analysts create, train, evaluate, and predict with machine learning models using SQL statements such as CREATE MODEL.
- Generative AI Studio
- A Vertex AI console and API for designing prompts, tuning Gemini models, and evaluating generative outputs without managing serving infrastructure.
- Document AI
- A Google Cloud service that extracts structured data from documents using pretrained or custom processors for OCR, forms, invoices, and identity documents.
- Vertex AI Agent Builder
- A low-code framework that combines Vertex AI Search and Conversation to create enterprise search and chat agents grounded in private data.
- Vertex AI Model Registry
- A central catalog that stores model versions, artifacts, and deployment metadata so teams can track, compare, and deploy models reproducibly.
- Vertex AI Experiments
- A tracking service that records parameters, metrics, and artifacts for each training run, enabling teams to compare experiments and reproduce results.
- Dataplex
- A data governance service that provides metadata discovery, data quality checks, lineage, and lifecycle management across BigQuery, Cloud Storage, and BigLake.
- Vertex AI Feature Store
- A managed repository that serves feature values online and offline, letting teams share engineered features and avoid inconsistent training-serving feature logic.
- IAM least privilege
- A security practice that grants each identity only the roles and permissions required for its task, reducing blast radius across shared projects and datasets.
- Vertex AI custom training
- A managed training service that runs a user-provided container on selected compute, allowing any ML framework and custom preprocessing code.
- Vertex AI Hyperparameter Tuning
- A service that runs multiple training trials with different hyperparameter values and identifies the best configuration by optimizing a chosen metric.
- Vertex AI Pipelines
- A managed orchestrator that runs containerized ML workflow steps as a directed acyclic graph and records artifacts and lineage in Vertex ML Metadata.
- Reduction server
- A Vertex AI distributed training option that aggregates gradients from multiple GPU workers to improve throughput for synchronous data-parallel training.
- Point-in-time correctness
- A feature engineering property where training rows use only feature values available before the label timestamp, preventing future information leakage.
- AutoML vs custom training
- A decision rule: use AutoML for standard data types when speed matters, and custom training when a framework, loss, or preprocessing step is unsupported.
- Vertex AI Endpoint
- A managed HTTP service that hosts one or more deployed model versions and returns online predictions with autoscaling and traffic splitting.
- Batch prediction
- A Vertex AI job that reads input from BigQuery or Cloud Storage, runs predictions without an endpoint, and writes output to a destination.
- Traffic splitting
- An endpoint feature that routes specified percentages of prediction requests to different model versions for canary releases or A/B testing.
- Private endpoint
- A Vertex AI endpoint that uses VPC peering and has no public IP, requiring network configuration and IAM for access.
- Prediction container
- A container image that implements the Vertex AI prediction HTTP contract to load a model and serve requests on an endpoint.
- Online vs batch serving
- A decision rule: use online endpoints for low-latency per-request predictions and batch prediction for high-throughput offline scoring jobs.
- Cloud Composer
- A managed Apache Airflow service for scheduling and monitoring workflows that span BigQuery, Vertex AI, Cloud Storage, and other Google Cloud services.
- Cloud Workflows
- A serverless orchestrator that executes YAML or JSON workflow definitions calling HTTP endpoints and Google Cloud APIs in sequence or parallel.
- Eventarc
- A service that routes events from sources such as Cloud Storage, Pub/Sub, and audit logs to triggers like Cloud Run and Workflows.
- Cloud Scheduler
- A fully managed cron service that invokes HTTP endpoints, Pub/Sub topics, or Workflows on a defined schedule.
- Vertex ML Metadata
- A metadata service that records artifacts, executions, and parameters produced by Vertex AI Pipelines to support lineage and reproducibility.
- Training-serving skew
- A difference between the feature distribution used during training and the distribution observed at serving time, often caused by inconsistent preprocessing.
- Prediction drift
- A change in the distribution of input features or predicted outputs over time relative to a baseline, indicating that model behavior may degrade.
- Vertex AI Model Monitoring
- A service that monitors deployed models for training-serving skew and prediction drift and writes results to BigQuery with optional alerts.
- Explainable AI attributions
- Vertex AI feature attribution methods such as Sampled Shapley and Integrated Gradients that estimate each input feature's contribution to a prediction.
- Cloud Monitoring alerting
- A Google Cloud service that evaluates metric conditions and sends notifications through channels such as email, SMS, Pub/Sub, or PagerDuty.