PMLE Exam Glossary - 65 Terms

Search the terminology pack for Professional Machine Learning Engineer. Use these definitions with the study guide and practice questions.

Download App Study Guide Free Practice Exam

A

A/B testing: A method for comparing different versions of a model to evaluate which performs better.
Apache Hadoop: A distributed data processing ecosystem mentioned as a data source and file type context for training and preprocessing.
Apache Spark: A distributed data processing framework used here for preprocessing, notebooks, and model development.
AutoML: A Google Cloud machine learning approach for training custom models using prepared data, including tabular workflows and forecasting models.

B

batch inference: Inference performed on data in batches rather than one request at a time.
bias: A risk in AI systems that can be monitored as part of responsible AI practices and solution readiness.
BigQuery: Google Cloud data warehouse used here for data exploration, preprocessing, and training data organization.
BigQuery ML: A Google Cloud capability for developing machine learning models directly in BigQuery, including model building, feature engineering or selection, and prediction generation.

C

CI/CD: Continuous integration and continuous delivery, used here for model deployment automation.
Cloud Build: Google Cloud’s build service used here for pipeline components and CI/CD model deployment.
Cloud Composer: Google Cloud’s managed orchestration service used for ML pipeline orchestration.
Cloud Run: Google Cloud’s serverless compute service used here for pipeline components and compute needs.
Cloud SQL: Google Cloud managed relational database service mentioned as a data source and as a minimum coding skill area for interpreting code snippets.
Cloud Storage: Google Cloud object storage used here for organizing and training on data.
Colab Enterprise: A Google Cloud Jupyter backend option for model prototyping.
containerized serving: Serving models inside containers as a scalable backend option.
CPU: A central processing unit, listed here as a compute option for training and serving models.

D

data lineage: The record of how data moves and changes across systems and processing steps.
Dataflow: Google Cloud’s data processing service used here for preprocessing and as a component in ML pipelines.
Dataproc: A Google Cloud service mentioned as a Jupyter notebook backend and as a serving/inference environment.
Document AI API: An industry-specific API mentioned as an example of an ML API used to build AI solutions.

E

Explainable AI: A set of techniques and tools for explaining model predictions and monitoring model behavior.

F

fairness: An AI readiness and monitoring concern related to equitable model behavior across groups.
feature attribution drift: A change over time in how input features contribute to model predictions.
feature engineering: The process of creating or transforming input variables to improve model training or prediction quality.
fine-tuning: The process of adapting a foundation model to a specific task or domain using additional training.
foundation models: Large pre-trained models used as the basis for generative AI solutions, which can be fine-tuned or used through APIs and tools such as Model Garden and Vertex AI Agent Builder.

G

generative AI: A class of AI systems that create new content or outputs, and in this guide includes building and evaluating solutions based on foundation models.
Google Cloud: Google’s cloud computing platform used here to build, train, deploy, monitor, and orchestrate machine learning and generative AI solutions.
GPU: A graphics processing unit, listed here as an accelerator option for training and serving models.

H

Horovod: A distributed training framework mentioned for training with TPUs and GPUs.
hyperparameter tuning: The process of searching for better model hyperparameter values to improve training or serving performance.

J

JAX: A machine learning framework mentioned as a common framework for developing models in Vertex AI Workbench.
Jenkins: A CI/CD automation tool mentioned for model deployment.

K

Kubeflow Pipelines: A pipeline orchestration framework used for ML development, experimentation, and end-to-end ML workflows.

M

ML APIs: Machine learning application programming interfaces used to build AI solutions, including APIs available from Model Garden and industry-specific APIs.
MLFlow: A third-party pipeline platform mentioned as something that can be hosted on Google Cloud.
MLOps: Foundational concepts for operationalizing machine learning, including training, deployment, monitoring, and lifecycle management of models.
model explainability: The ability to understand and explain why a model produced a particular prediction.
Model Garden: A Google Cloud offering that provides machine learning APIs and access to foundation and open-source models for building AI solutions.
model lineage: The record of a model’s origin, transformations, and relationships across its lifecycle.

O

online inference: Inference served interactively in response to requests, as opposed to batch processing.

P

PHI: Abbreviation for protected health information, treated here as sensitive data with privacy implications.
PII: Abbreviation for personally identifiable information, treated here as sensitive data with privacy implications.
PyTorch: A machine learning framework used for model development, experiments, training, and serving.

R

RAG: Abbreviation for retrieval augmented generation.
responsible AI: A set of practices for building AI systems that consider issues such as bias, fairness, and safe use.
Retail API: An industry-specific API mentioned as an example of an ML API used to build AI solutions.
retrieval augmented generation: An application pattern in which a system uses Vertex AI Agent Builder to implement retrieval-augmented generation (RAG) applications.

S

sklearn: A machine learning library mentioned as a common framework for developing models in Vertex AI Workbench.
Spanner: Google Cloud’s globally distributed database mentioned as a data source for organization-wide data.

T

Tabular Workflows: A feature of AutoML mentioned in the context of preparing data for tabular model training.
TensorFlow: A machine learning framework used for model development, experiments, and training.
TensorFlow Extended: A TensorFlow-based platform for building production ML pipelines and preprocessing workflows.
TFX: Abbreviation for TensorFlow Extended.
throughput: A performance measure used here to scale the serving backend and evaluate model serving capacity.
TPU: A tensor processing unit, listed here as an accelerator option for training and serving models and for distributed training.
training-serving skew: A mismatch between the data or preprocessing used during training and the data or preprocessing used during serving.

V

Vertex AI Experiments: A Google Cloud environment for tracking and running ML experiments and comparing model artifacts and versions.
Vertex AI Feature Store: A Vertex AI service for creating, consolidating, and serving features for machine learning workflows.
Vertex AI Model Monitoring: A Vertex AI service for continuously evaluating models and monitoring for issues such as training-serving skew and feature attribution drift.
Vertex AI Prediction: A Vertex AI serving option used for online model prediction and explainability-related use cases.
Vertex AI TensorBoard: A Vertex AI tool used with TensorFlow and PyTorch for development and experimentation.
Vertex AI Workbench: A Google Cloud Jupyter notebook environment used for prototyping and model development.
Vertex ML Metadata: A Vertex AI metadata service used for tracking and comparing model artifacts and versions.

About These Definitions

These definitions are loaded from the shared release pack. Use them with the study guide and practice questions to connect vocabulary to exam scenarios.

Download App Read the full study guide Take the free practice exam