Driving collaboration with privacy-preserving computation
April 1, 2021
April 1, 2021
It’s no secret that data drives intelligent decision-making. Across industries, companies with superior data and analytics practices outperform their competitors. It also shouldn’t be a surprise that companies can unlock even more value by sharing data with each other. Gartner has predicted that by 2023 – just two years from now! – organizations that promote data sharing will outperform their peers on most business metrics.
But most companies are wary of collaborating when it comes to sharing data. Sometimes they’re worried about preserving their intellectual property. Other times they aren’t sure how to collaborate without running afoul of regulations like GDPR or HIPAA, which can prevent direct sharing of data. Whatever the reason, companies are leaving a lot of value untapped.
That’s why we’re building on our previous work with privacy-preserving technologies to develop a new framework for privacy-preserving data cooperatives. With a well-designed cooperative, companies can share data and collaborate without concerns around trust, compliance, privacy and data control or ownership. There are many technological techniques available to make this possible; confidential computing, homomorphic encryption and multiparty computation can all contribute. But most important, we’re working to standardize the interfaces between organizations, making it easy for them to join a cooperative effort.
Given the challenges of the last year, the idea of data sharing has received particular interest from healthcare. But in the healthcare industry, perhaps more than anywhere else, maintaining data privacy is essential. Patients’ medical records must be protected. Enter privacy-preserving data cooperatives! If we can deliver ways to share data between health institutions while maintaining patient privacy, AI and other tools can help expand knowledge by analyzing data from multiple sources.
We built a proof of concept to show the value of our framework for a very practical case: detecting sepsis. Sepsis is a life-threatening condition that can occur as the body responds to infection. It’s a serious challenge for hospitals. It’s easy, then, to understand why hospitals would be eager for a better way to detect it. And by getting multiple hospitals to participate in a data cooperative, we can build a more accurate sepsis detection model.
We evaluated different techniques to bring this scenario to life. Ultimately, we built a proof of concept that uses a combination of Intel SGX secure enclave technology, enabled by the Fortanix Confidential Computing Manager, along with federated learning. It demonstrates how to train and use an AI model to detect sepsis in just this way.
Individual models are trained at the individual participants’ ends – that is, at the participating hospitals. The resulting models are then aggregated inside a secure enclave. This creates a single, global and “more knowledgeable” AI model without any hospital directly sharing its data with any other hospital, nor with the cooperative itself!
Each hospital would then benefit from a globally trained model to be used for their local “predictions” on their private patients’ datasets. If they used a model trained with this approach, individual hospitals could apply the model to real-time data from their own current patients in order to better detect cases of sepsis early, when there’s a better chance to treat it. The cooperative framework approach itself can also be reused by the participants for different, future use cases with different data sets. They can continue to build collaborative insights while maintaining their data privacy.
The proof of concept is built as a prototyped extension of Accenture Applied Intelligence’s AIP+ service, a collection of modular, pre-integrated AI services and capabilities designed to simplify the adoption of AI solutions.
Using privacy preserving data cooperatives guarantees that all participants will have a standardized infrastructure to share. It also means they can stay in compliance with data regulations and maintain their own data privacy while using it. As these cooperatives become more established, they will also offer standardized on-boarding: new participants can join easily, encouraging continued and expanded collaboration.
Data privacy concerns are always top of mind for enterprise. Rightly so! But with the evolution of privacy-preserving and confidential computing techniques, companies can reap the benefits of data sharing for garnering richer and deeper comprehensive intelligence without sharing the data itself.
Watch this space for more about our efforts in privacy-preserving computation. For more information or to schedule a demo, email Giuseppe Giordano and Kuntal Dey.