AWS clean-rooms documentation change
Summary
Restructured documentation about custom ML modeling workflow, added detailed technical implementation steps, monitoring details, and results management
Security assessment
Added details about secure processing within AWS Clean Rooms collaboration environment, access control for results, and monitoring through CloudWatch. These describe security features but do not indicate a specific security vulnerability being addressed.
Diff
diff --git a/clean-rooms/latest/userguide/working-with-custom-models.md b/clean-rooms/latest/userguide/working-with-custom-models.md index 4020a0d63..1a6cc3b8f 100644 --- a//clean-rooms/latest/userguide/working-with-custom-models.md +++ b//clean-rooms/latest/userguide/working-with-custom-models.md @@ -5,2 +4,0 @@ -Next steps - @@ -13 +11,21 @@ From a technical standpoint, the following diagram describes how custom ML model - 1. Package your models (training or inference) in a container image and publish to Amazon ECR. +Here's how custom ML modeling works in Clean Rooms ML: + + 1. Data Source Configuration + + * Source data can be stored in Amazon S3 catalog, in the AWS Glue Data Catalog, or Snowflake + + * AWS Glue Data Catalog is used to organize and catalog + + * Data from multiple AWS accounts can be used within the same collaboration + + 2. SQL Query and Data Processing + + * SQL queries are used to access and process the source data + + * The queries run within the AWS Clean Rooms collaboration boundaries + + * Processed data feeds into ML Input Channels for model training + + 3. ML Model Development + + * Source code for the model can be developed using AWS Deep Learning Container Images @@ -15 +33 @@ From a technical standpoint, the following diagram describes how custom ML model - 2. Create the AWS Clean Rooms and Clean Rooms ML resources needed to perform model training. + * Custom container images must be created and stored in Amazon Elastic Container Registry @@ -17 +35 @@ From a technical standpoint, the following diagram describes how custom ML model - 3. Associate the model algorithm to the collaboration. + 4. Infrastructure Components @@ -19 +37 @@ From a technical standpoint, the following diagram describes how custom ML model - 4. Read the data from the data provider accounts to generate the ML input channel that is used for training or inference. + * Amazon Elastic Container Registry stores and manages the ML model containers @@ -21 +39 @@ From a technical standpoint, the following diagram describes how custom ML model - 5. Run the ML training job with the information from steps #1 and #4. + * ML processing occurs within the secure AWS Clean Rooms collaboration environment @@ -23 +41 @@ From a technical standpoint, the following diagram describes how custom ML model - 6. (Optional) Export the trained model artifacts to the results receiver. + 5. Monitoring and Logging @@ -25 +43,9 @@ From a technical standpoint, the following diagram describes how custom ML model - 7. (Optional) Run the ML inference job with the information from Steps #1, #4, and #5. + * Amazon CloudWatch provides metrics and logs for both collaborating parties + + * Monitoring is available across AWS accounts involved in the collaboration + + * Performance metrics and operational logs are accessible to relevant parties + + 6. Results Management + + * Access to results is controlled according to collaboration permissions @@ -50,11 +75,0 @@ Before you get started, see the [Custom ML modeling prerequisites](./custom-mode - * Next steps - - - - -## Next steps - -After you have created a custom model, you are ready to: - - * [Create a collaboration and membership in AWS Clean Rooms](./working-with-collaborations.html) -