Master's thesis; AI-enabled Data Privacy Protection

About us
RISE Research Institutes of Sweden AB is a research organization owned by the Swedish government. This thesis will be conducted within the Cybersecurity Unit, which is among the largest public sector cybersecurity research groups in Sweden. Our core areas of expertise are: IoT Security, Cloud Security, Network & Communication Security, Access Control, Privacy (technical and social aspects), and Secure Virtualization and Trusted Computing. The RISE Cybersecurity Unit is the European leader in IoT security research & development. In additional to a strong research environment, RISE Cybersecurity is the owner of the RISE Cyber Range, a cybersecurity test and demo facility in Kista with a critical infrastructure grade security, that provides a trusted place for Swedish industry to understand and address their cybersecurity needs. RISE Cyber Range, in addition to providing practical cybersecurity education, training and exercise, is an environment for state-of-the-art cybersecurity research and development.

Background
Modern machine learning (ML) analytics rely on the availability of large volume and diverse data that can be used to train robust and reliable ML algorithms. However, data sharing for ML analytics is often limited due to the privacy concerns of the individuals whose information is harvested and data protection regulations such as GDPR that prohibit sensitive data sharing. To this end, it is common for data holders to apply privacy-enhancing transformations such as aggregation, hiding, sampling, and perturbation, on their datasets before sharing them for ML analytics. However, such transformations may destroy the utility of the data for downstream ML analytics or provide limited privacy protection against informed adversaries.

Thesis description
In this project, we will investigate how artificial intelligence (AI), in particular reinforcement learning, can help data owners to automatically protect their sensitive data against privacy inference attacks while retaining sufficient utility for ML analytics. To this end, we will design an adversarial learning environment within which an autonomous agent that performs privacy-enhancing transformations on sensitive data can simultaneously interact with privacy attackers and utility functions to discover novel approaches for sharing a privacy- and utility-preserving version of the dataset for downstream ML analytics.

RISE will provide background information and the necessary guidance during the master thesis work. The tasks of the student for this master thesis project are:

• Study state-of-the-art privacy-inference attacks against data sharing and the corresponding approaches for protection.
• Familiarize with reinforcement learning concepts and algorithms.
• Design and implement a learning environment encompassing privacy adversaries, utility functions, and an autonomous agent that learns to privatize the sensitive data by employing defensive actions.
• Experimentally evaluate the correctness and efficiency of the developed environment on various sensitive data sharing applications.
• Document the activities and results as a thesis report.

Student profile
We are looking for an ambitious MSc student who has fulfilled the course requirements. Good Python programming skills are required, as is good spoken and written English. Experience with machine learning, reinforcement learning, and privacy issues in data sharing is a plus. 

Welcome with your application
For more information, please contact Apostolos Pyrgelis (apostolos.pyrgelis@ri.se). Last day of application is November 30, 2024. Applications should include a brief personal statement, a CV, and a list of grades. In the application, make sure to mention previous activities or other projects that you consider relevant for the position. Candidates are encouraged to send in their application as soon as possible. Suitable applicants will be interviewed as applications are received.