Motivation & Goal
Several useful services that use machine learning (ML) algorithms to categorize and classify large amounts of sensitive data have emerged in the recent past. However, in order to use these services, current solutions require the disclosure of personal information. As a result, there is an inherent conflict between utility and privacy: ML requires data to operate, whereas privacy requires sensitive information to be kept private. This sparked the field of Privacy-Preserving Machine Learning (PPML), which ensures data privacy. The goal of the position is to work on PPML challenges such as:
- Design and implementation of a PPML protocol for various settings using Secure multi-party computation (MPC). The settings include two, three and four server cases as well as passive and active security models.
- Implementation of the designed PPML framework (consisting of model training and inference) in C/C++ using the MOTION framework .
- Comparisons with state-of-the-art ML models in plaintext primarily using the PyTorch framework.
The results emerging from this work are essential contributions to research papers that will be published at international top conferences.
 Lennart Braun, Daniel Demmler, Thomas Schneider and Oleksandr Tkachenko. MOTION – A Framework for Mixed-Protocol Multi-Party Computation. ACM TOPS'22
- Good programming skills in C/C++ and Python/PyTorch.
- At least basic understanding of machine learning and cryptography.
- High motivation and creativity + ability to work independently.
- Flexible working hours.
- Experience with reading research papers; Knowledge of the English language goes without saying.
If you are interested, please get in touch and send your application (including a CV and transcript of records) to: