What's changed: Added per-section figures (cert-figure-retrofit). New AI-300 Chapter 1 (Domain 1 "MLOps infrastructure": workspace/datastores/compute targets/IAM; asset management = data assets/environments/components/registries; IaC and automation = Bicep-Azure CLI/GitHub Actions/workload identity federation/network restriction/Git)
1.2Managing assets in Machine Learning
Understand creating and managing AML assets (data assets, environments, components) and sharing assets across workspaces using registries, for reproducibility and reuse.
The keys to MLOps are reproducibility and reuse. AML manages these as assets (data, environments, components, models) that can be versioned and shared.
1.2.1Data assets, environments, and components
- Data asset: a versioned reference to specific data (file/folder/table), making training inputs reproducible.
- Environment: defines dependencies (Python packages, Docker) needed to run, so jobs are reproducible.
- Component: a reusable step bundling inputs, outputs, and logic—combined as building blocks of a pipeline.
1.2.2Sharing via registries
A registry shares models, environments, components, and data assets across multiple workspaces. For example, a component built in a dev workspace can be reused from a production workspace. This propagates standardized building blocks across the org and improves MLOps consistency.
| Asset | Role | Reproducibility/reuse |
|---|---|---|
| Data asset | Versioned reference to data | Reproducible inputs |
| Environment | Dependencies (packages/Docker) | Reproducible runtime |
| Component | Reusable processing step | Reuse in pipelines |
| Registry | Cross-workspace sharing | Standardize/propagate org-wide |
Watch the mix-ups: (1) environment (dependency definition) vs compute (the machine that runs it). (2) Combine components (reusable steps) to build a pipeline. (3) A registry (cross-workspace sharing) is a separate layer from in-workspace asset management.
Map requirement → asset: "make training input reproducible" = data asset; "pin dependencies for reproducibility" = environment; "componentize logic for reuse" = component; "share building blocks across workspaces" = registry.
1.2.3Section summary
- Assets = data asset (reproducible input) / environment (reproducible dependencies) / component (reusable step)
- Registries share and standardize assets across workspaces
- Reproducibility and reuse are the core of MLOps
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. Which is a versioned reference to data that makes training inputs reproducible?
Q2. Which defines dependencies like Python packages and Docker so jobs are reproducible?
Q3. Which unit bundles inputs, outputs, and logic for reuse as a building block of a pipeline?
Q4. Which is used to share models and components across multiple workspaces?
Q5. Which correctly distinguishes an environment from compute?
Q6. Which is a basic principle to improve reproducibility and reuse in MLOps?