Machine learning infrastructure is the software, hardware, and services that make data processing, model training, deployment, serving, monitoring, and maintenance repeatable and reliable. It supports data scientists, data engineers, MLOps engineers, platform engineers, and AI applications across the development cycle. section 1.1.2
A reference infrastructure stack commonly includes:
- Data warehouses and ingestion pipelines for collecting, transforming, enriching, labeling, and processing data. section 1.1.2
- Compute platforms, workflow orchestration services, and training services for executing development workloads. section 1.1.2
- Model metadata and artifact stores, serving services, and monitoring systems for deployment and production operation. section 1.1.2
Its central production requirement is a smooth path from data and experimentation to scalable deployment and monitoring. Automation reduces operational work, communication, and coordination costs, allowing data scientists to complete more of the cycle independently without becoming DevOps or data-engineering specialists. section 1.1.4
Productionizing a model
A data scientist uses shared infrastructure to set up a data pipeline, train a model, deploy it for production use, and monitor model performance, while the platform supports the surrounding automation and operations. section 1.1.2 section 1.1.4
