PMLE cheat sheet

Exam at a glance

Domain weightings

How much of the exam each domain covers. Spend your study time in proportion — the heavier the domain, the more questions you'll see.

Key terms

Vertex AI

Vertex AI is Google Cloud's unified platform for building, training, deploying, and managing machine learning and generative AI models. PMLE covers it across the model-development, serving, and pipeline domains.

BigQuery ML

BigQuery ML is a capability that lets you create and run machine learning models directly in BigQuery using SQL. PMLE covers it for architecting low-code AI solutions.

AutoML

AutoML is a set of Google Cloud capabilities that train high-quality models with minimal code by automating architecture and hyperparameter search. PMLE covers it for low-code AI solutions.

Model Garden

Model Garden is a catalog on the Gemini Enterprise Agent Platform for discovering, testing, and deploying foundational and open models. PMLE covers it for building AI solutions from foundational models.

Gemini

Gemini is Google's family of multimodal foundational models available on Google Cloud for building generative AI solutions. PMLE covers selecting and adapting such models in the low-code AI domain.

Feature Store

Vertex AI Feature Store is a managed service for storing, serving, and sharing machine learning features consistently across training and serving. PMLE covers it for managing data and models across teams.

Vertex AI Pipelines

Vertex AI Pipelines is a service for orchestrating machine learning workflows as reproducible, containerized steps built on Kubeflow Pipelines. PMLE covers it in the automating-and-orchestrating domain.

Kubeflow

Kubeflow is an open-source platform for building portable, scalable machine learning workflows on Kubernetes. PMLE covers Kubeflow Pipelines as a foundation for ML pipeline orchestration.

Tensor Processing Unit

A Tensor Processing Unit (TPU) is a Google-designed accelerator optimized for large-scale machine learning training and inference. PMLE covers choosing between CPUs, GPUs, and TPUs when scaling models.

Distributed training

Distributed training is the practice of splitting model training across multiple devices or machines to handle large models and datasets. PMLE covers it in the scaling-prototypes-into-ML-models domain.

Online prediction

Online prediction is low-latency, real-time model serving through an endpoint that responds to individual requests. PMLE covers it, alongside batch prediction, in the serving-and-scaling domain.

Batch prediction

Batch prediction is model inference run over a large set of inputs at once, without low-latency requirements. PMLE covers it and online prediction when choosing a serving strategy.

Training-serving skew

Training-serving skew is a discrepancy between how features are computed during training and during serving, which degrades model performance. PMLE covers detecting it when monitoring AI solutions.

Model drift

Model drift is the degradation of a model's performance over time as the statistical properties of production data change. PMLE covers monitoring for drift and skew in the monitoring domain.