A new Federated Learning project could solve GDPR-issues

In contrast to ordinary machine learning where all the data is shared and centralized to train one model, federated machine learning instead splits its model training to run on all the participating actors locally stored sensitive data in parallel, without letting any one actor see the data or get access to it. The result of the individually trained models is aggregated to one main working model, which in the end becomes as good as an ordinary machine learning model, or even better.

This new software technology has the promise to enable organizations to share access to their data, without the other party seeing it. This is a big step towards managing GDPR-issues.

“The support from the industry and the public sector is tremendous and many actors can relate to the challenges we describe concerning sharing data, both within and outside their organizations” says Erik Wilson, Project Manager at AI Sweden.

Virtual testbed

AI Sweden is launching a virtual testbed within this new project together with partners including Zenuity, Scaleout, RISE, Peltarion, MAQS, Qamcom, HPE, and Fraunhofer Chalmers Research Centre for Industrial Mathematics (FCC). 

The project will be ongoing for eight months and the goal of the project is to build up a collective know-how of how to work with this technology in a production setting, while also having created a legal framework to speed up the process for future partners that would like to explore federated learning. The project is finished in march 2021.

Facts
Partners: AI Sweden, Zenuity, Scaleout, RISE, Peltarion, MAQS, Qamcom, HPE, and Fraunhofer Chalmers Research Centre for Industrial Mathematics (FCC). Funding from Vinnova.

 

What to know more about the project Federated learning? 

Project Manager
Erik Wilson

erik.wilson@ai.se 
+46 (0)70-425 90 70