About This Flashcard Deck
This flashcard deck contains 10 cards covering key ML Engineer concepts for the PMLE exam. Test your GCP ML engineering knowledge. Use active recall by attempting to answer each question before revealing the answer.
Question
What is Vertex AI?
Click to reveal answer
Answer
Unified ML platform for building, training, and deploying models. Includes AutoML, custom training, Feature Store, Pipelines, and Prediction.
Click to flip back
All ML Engineer Flashcards
Q: What is Vertex AI?
A: Unified ML platform for building, training, and deploying models. Includes AutoML, custom training, Feature Store, Pipelines, and Prediction.
Q: When to use AutoML vs custom training?
A: AutoML: limited ML expertise, quick iteration, structured data/images/text. Custom: need full control, specific architectures, research.
Q: What is Vertex AI Feature Store?
A: Managed repository for ML features. Ensures consistent features between training and serving. Supports batch and online serving.
Q: What is model drift?
A: When model performance degrades because the data distribution changes over time. Two types: data drift (input changes) and concept drift (relationship changes).
Q: What is Vertex AI Pipelines?
A: Managed service for orchestrating ML workflows. Supports Kubeflow Pipelines and TFX. Tracks lineage and artifacts.
Q: What is hyperparameter tuning?
A: Automatically searching for the best hyperparameters (learning rate, batch size, layers). Vertex AI Vizier uses Bayesian optimization.
Q: What are pre-trained APIs?
A: Ready-to-use ML APIs: Vision AI, Natural Language AI, Translation AI, Speech-to-Text, Video Intelligence. No training needed.
Q: What is TFX?
A: TensorFlow Extended — production ML pipeline framework. Components: ExampleGen, StatisticsGen, SchemaGen, Transform, Trainer, Evaluator, Pusher.
Q: What is Explainable AI?
A: Vertex AI feature that provides feature attributions — showing which features most influenced a prediction.
Q: What is the ML training-serving skew?
A: When the data processing in training differs from serving, causing inconsistent predictions. Feature Store helps prevent this.
GCP Flashcard Study Approach
Google Cloud exams emphasise service selection and architecture decisions. Use these flashcards to build instant recall of GCP service capabilities, then apply that knowledge to scenario-based practice questions. Pay special attention to cards about managed vs. unmanaged services and serverless options — GCP strongly favours managed and serverless architectures in their exam scenarios.