AWS Clean Rooms supports synthetic dataset generation training custom ML training

AWS Clean Rooms now enables you and your partners to generate privacy-enhancing synthetic datasets from your collective data to train regression and classification machine learning (ML) models.

Synthetic dataset generation allows you and your partners to create training datasets with similar statistical properties to the original data, without the training code having access to real records. This new capability de-identifies subjects—such as people or entities about whom data has been collected—in the original data, mitigating the risk that a model will memorize information about individuals in the training data. This unlocks new ML model training use cases that were previously restricted by privacy concerns, such as campaign optimization, fraud detection, and medical research. For example, an airline with a proprietary algorithm wants to collaborate with a hotel brand to offer joint promotions to high-value customers, but neither organization wants to share sensitive consumer data. Using AWS Clean Rooms ML, they can generate a synthetic version of their collective dataset to train the model without exposing raw data—enabling more accurate promotions targeting while protecting customer privacy.

For more information about the AWS Regions where AWS Clean Rooms ML is available, see the AWS Regions table. To learn more, visit AWS Clean Rooms ML.