Privacy Preserving Machine Learning for Industrial Applications (PRIMAL)

Term: 1/2020 - 12/2022 (36 Monate)

Partner: SCCH, Hagenberg

Topic:
PRIMAL will enable industrial deep learning applications by increasing the amount of usable data sources. This will be achieved by developing privacy preserving deep transfer learning methods. Privacy preservation allows utilizing data from even commercially competing parties while by means of transfer and multi-task learning data from different (but related) sources can be leveraged. The result will be a software framework providing algorithms and interfaces to build privacy preserving predictive analytics applications for a wide range of (industrial) applica-tions, exemplified in the independent areas of intralogistics, welding technology and bioinfor-matics. There are many open issues beyond the state-of-the-art to be solved to reach this goal: protection against recently devised privacy attacks; evaluation of methods beyond image classi-fication tasks, i.e. suitability of approaches for industrial applications; support for transfer and multi-task learning as well as recurrent networks; interfaces for integration into (IoT) data anal-ysis infrastructures including authentication concepts.