A12 - registries and cross-workspace asset sharing

• organisations may have many AML workspaces using the same ML assets
• an Azure ML registry provides a shared location for ML assets that can be used across multiple workspaces

• the same versioned assets can be reused where required, aiding in reproducibility

• but shared availability also does not mean that the latest version must be used everywhere, ie. no automatic promotion

• cross-workspace collaborationgives teams a controlled boundary between experimentation and reusable organizational assets

EXPERIMENTATION
workspace
    ↓
validated asset
    ↓
SHARED REGISTRY
    ↓
downstream workspaces

• example architecture:

				 GITHUB
					│
					▼
			  AML-RESEARCH
					│
			 train / evaluate
					│
			   MLflow tracking
					│
					▼
			 approved assets
					│
					▼
              ORGANIZATION REGISTRY
		 ┌──────────┼───────────┐
		 ▼                              ▼                                ▼
	   model                environment             component
		 │                             │                                │
		 └──────────┼───────────┘
			┌───────┴────────┐
			▼                                                ▼
	   AML-STAGING                      AML-PRODUCTION

• security still applies, so use least privilege based authentication and authorisation