Amazon’s launching (in preview) a privacy-preserving service that lets AWS customers deploy “lookalike” AI models trained for one-off company-company collaborations.
Called Clean Rooms ML — an offshoot of AWS’ existing Clean Rooms product — the service removes the need for AWS customers to share proprietary data with their outside partners to build, train and deploy AI models.
“With Clean Rooms ML, you can train a private lookalike model across your collective data,” Swami Sivasubramanian, VP for data and machine learning services at AWS, said onstage during a keynote this morning at AWS re:Invent. “You can keep control of your models and delete them when you’re done.”
Clean Rooms ML allows customers to take a small sample of customer records to generate an expanded set of similar records with a partner — a lookalike AI model. For example, an airline might take signals about loyal customers and work with an online booking service to offer promotions to new, but very similar, users.
Clean Rooms ML offers controls to tune model outputs based on particular business needs, Sivasubramanian said. In the near future, he added, AWS plans to add settings specifically for applications in healthcare.
In a related announcement, Amazon unveiled Clean Rooms Differential Privacy, a fully managed service in Clean Rooms. Clean Rooms Differential Privacy obfuscates the upload of any customer’s data while generating “aggregate insights,” Amazon says — allowing customers to get combined insights about things like advertising campaigns, investment decisions and clinical research without having to expose their proprietary data.
