Contribute Media
A thank you to everyone who makes this possible: Read More

BigMedilytics: MPyC in practice


In this talk we would like to introduce secure Multi-Party Computation (MPC) - as it allows for privacy preserving analytics - and show the possibilities of the framework MPyC. By utilizing MPyC, TNO, together with an insurance company and a hospital, have built a solution that can predict the number of hospitalization days on datasets from both parties, without them sharing their data.

TNO together with an insurance company and a hospital are analyzing data of heart failure patients. The insurance company has a lot of data on medical compliance and claim behavior, whereas the hospital has data from an academic study of hundreds of people that we tracked throughout several years. The hospital has data among others about exercising performance, alcohol intake, smoking behavior, etc. history. Combining these datasets can create more insights than analysis on each dataset separately. However, given the privacy concerns, this is not feasible. This is where Multi-Party Computation (MPC) offers a solution. MPC comprises of a set of cryptographic techniques that enable analysis in the encrypted domain. With these techniques it can be mathematically shown that no sensitive data is leaked, when performing an analysis. MPyC, developed by Berry Schoenmakers (TU/e), is a python library that is an MPC framework based on the technique secret sharing. By using this library, we have been able to successfully implement a LASSO regression.


Improve this page