AWS Clean Rooms launches privacy-enhancing synthetic dataset generation for ML model training

This article announces privacy-enhancing synthetic dataset generation for AWS Clean Rooms, enabling organizations to generate synthetic training datasets from collective data while protecting individual privacy.

• Generate synthetic datasets preserving statistical patterns without exposing original records
• Addresses tension between data utility and privacy protection in ML model training
• Uses advanced ML techniques to de-identify subjects while maintaining data characteristics
• Mitigates re-identification risk through model capacity reduction techniques
• Organizations control privacy parameters including noise levels and membership inference attack protection
• Provides fidelity and privacy quality metrics using KL-divergence and protection scores
• Integrated into AWS Clean Rooms ML workflow with enhanced analysis templates
• Supports classification and regression models on tabular data
• Billed separately based on Synthetic Data Generation Units (SDGUs)
• Available in all AWS regions where Clean Rooms is available

This capability enables secure collaborative ML model training on sensitive data while quantifiably reducing re-identification risks and maintaining compliance requirements.