SMiLe - Secure Machine Learning Applications with Homomorphically Encrypted Data

The project investigates the conditions under which solutions using homomorphically encrypted data can help exploit sensitive data for machine learning applications. The focus is on the further development of relevant software components as well as the transfer of relevant know-how.

Short Description

Companies increasingly depend on their ability to leverage data to improve their efficiency and create innovative services in order to succeed. Data must be combined across organisational units and beyond company boundaries if the enterprise is to profit from the opportunities offered by machine learning. Data sharing is difficult, however, especially when it comes to sensitive data.

Legislation such as the GDPR requires organisations to take appropriate security measures, and essentially restricts the use and sharing of data. Companies also have a strong interest in protecting their intellectual property and are therefore reluctant to share their data. Given this situation, it is not surprising that the potential offered by the data that companies gather and store remains largely unexploited. The researchers involved in the SMiLe project aim to show how, in the future, this potential can be harnessed to a greater extent.

So-called 'homomorphous encryption' offers a promising approach to achieving this goal. Data encrypted in this manner are protected from unauthorised access, but can still be used for computations. The results of these computations are also encrypted and thus protected. To date, this approach has mainly been explored in theoretical studies, with the practical expertise and appropriate software still lacking. By addressing both these factors, SMiLe will create an essential basis for the practical use of machine learning on encrypted data.

Two use cases will assess the potential of this approach for machine learning: 

  1. workforce segmentation and 
  2. predictive machine maintenance. 

The project will address technical as well as social, legal and economic issues. The researchers will evaluate the solution approaches in terms of their analytical capabilities, performance and scalability, as well as their cost-effectiveness, transparency and ease of use. The advantages and disadvantages of machine learning using homomorphically encrypted data will be compared to alternative approaches.

SMiLe therefore helps to establish a cooperative-creative ecosystem in which different actors interact in a trustful, symbiotic and autonomous manner, realising previously unimaginable solutions which not only guarantee privacy and security, but also exploit hitherto untapped data potentials.

Project Partners

Consortium lead

Fraunhofer Austria Research GmbH

Other consortium partners

  • MCI Management Center Innsbruck Internationale Hochschule GmbH
  • Software Competence Center Hagenberg GmbH
  • VRVis Zentrum für Virtual Reality und Visualisierung Forschungs-GmbH
  • CORE smartwork GmbH
  • Fill GmbH, Tributech Solutions GmbH

Contact Address

Fraunhofer Austria Research GmbH
Dr. Daniel Bachlechner
Theresianumgasse 7
A-1040 Vienna