Doctoral student in Computer Science with focus on privacy-preserving neural models

Subject description

Despite GDPR claims on pseudonymization as a key privacy protection method, there is insufficient evidence from research that we can trust pseudonymization as a bulletproof method of personal identity protection. This PhD project will experimentally investigate the effectiveness of pseudonymization techniques in protecting writer identity across various natural language processing applications, from written essays to Robotic/AI interactions.

The research challenges focus on the fundamental question: Does pseudonymization work as we expect it to? This project will systematically evaluate different pseudonymization strategies through rigorous re-identification attacks to determine their true effectiveness in preserving privacy.

The research is structured around two interconnected themes:

Pseudonymization Effectiveness Analysis: Systematically evaluating how different identifiers, their number, and combinations affect privacy protection levels in text data. The project will assess which pseudonymization techniques provide adequate protection and under what circumstances they may fail.

Advanced Re-Identification Attack Methods: Developing and employing traditional attacks (anonymization reversal, information aggregation), recent techniques (graph/node attacks), and novel attacking methods to comprehensively test pseudonymization robustness in both static text (essays, documents) and interactive systems (conversational AI, robotics).

You will employ a methodological approach comprising three types of re-identification studies, each focusing on different categories for pseudonymization, their number, and their combination - in relation to the provided levels of protection. You will conduct motivated intruder tests to assess the likelihood of re-identification risks in various contexts, with special attention to written essays and human-robot interaction scenarios where personal information may be exposed.

This multidisciplinary approach will allow you to evaluate pseudonymization effectiveness across different modalities and interaction types, creating a comprehensive framework for privacy protection in AI systems.

Description of the workplace

The position is at Lund University's Department of Computer Science within the research division Robotics and Semantic Systems (RSS). The RSS focuses on research and teaching in AI, Machine Learning, Robotics and Robotic Learning, Human-Robot Interaction, and Natural Language Processing. Together with the Department of Automatic Control, the division operates RobotLab LTH, which gives access to many types of robots.

You will be working in a team with other PhD students and researchers at RSS, VR Project, and WARA ML, who form a multidisciplinary team working at the intersection of machine learning, security, and robotics. The position offers opportunities for international collaboration with leading research institutes and technology companies at the forefront of AI development.

More information about the Department of Computer Science (cs.lth.se)
More information about RSS (rss.cs.lth.se)
More information about the RobotLab LTH (robotics.lth.se)
More information about VR Project (mormor-karl.github.io)
More information about WARA ML (waraml.org)

Work duties

The main duties of doctoral students are to devote themselves to their research studies, which includes participating in research projects and doctoral courses. The work duties also include teaching and other departmental duties (up to 20%).

In addition to these general duties, you are expected to:

• Conduct research on privacy-preserving techniques for LLMs in robotic applications
• Develop and evaluate methods to detect and mitigate re-identification risks
• Participate in regular research seminars, workshops, and international conferences
• Collaborate with industry partners on real-world applications and case studies

Admission requirements

A person meets the general admission requirements for third-cycle courses and study programmes if the applicant:

• Have been awarded a second-cycle qualification, or
• Have satisfied the requirements for courses comprising at least 240 credits, of which at least 60 credits were awarded in the second cycle, or
• Have acquired substantially equivalent knowledge in some other way in Sweden or abroad

A person meets the specific admission requirements for third cycle studies in Computer Science if the applicant has:

• at least 60 second-cycle credits at an advanced level with relevance for the research topic, or 
• an MSc in Engineering in Computer Science and Engineering, Electrical Engineering, Information and Communication Technology, Engineering Physics or Engineering Mathematics. 

Additional requirements:

• Very good oral and written proficiency in English
• Knowledge in machine learning and deep learning, especially transformer architectures
• Knowledge of privacy-preserving techniques (differential privacy, federated learning, etc.)
• Programming experience in Python and relevant ML frameworks (PyTorch, TensorFlow)
• Experience with robotics platforms and frameworks (ROS, etc.) 
• Understanding of privacy and security principles in AI systems

Assessment criteria

Selection for third-cycle studies is based on the student’s potential to profit from such studies. The assessment of potential is made primarily on the basis of academic results from the first and second cycle. Special attention is paid to the following: 

1. Knowledge and skills relevant to the thesis project and the subject of study. 
2. An assessment of ability to work independently and to formulate and tackle research problems. 
3. Written and oral communication skills.
4. Other experience relevant to the third-cycle studies, e.g. professional experience.

Consideration will also be given to good collaborative skills, drive and independence, and how the applicant, through their experience and skills, is deemed to have the abilities necessary for successfully completing the doctoral program.

We offer

Lund University is a public authority which means that employees get particular benefits, generous annual leave and an advantageous occupational pension scheme. Read more on the University website about being a Lund University employee Work at Lund University. 

Terms of employment

Only those admitted to doctoral studies may be appointed to a doctoral studentship. The position is a full-time, fixed-term employment of a maximum of 5 years (including 20% departmental duties). Doctoral studentships are regulated in the Higher Education Ordinance (1993:100), chapter 5, 1-7 §§.

How to apply

Applications should be written in English and include:

• A personal letter explaining your interest in the position and how your background aligns with the research themes (max 2 pages)
• Curriculum vitae
• Degree certificate or equivalent
• A brief research statement outlining your ideas on privacy-preserving LLMs for natural texts, Robotics/AI interactions (max 2 pages)
• Contact information for at least two references
• Relevant publications or writing samples (if available)

Welcome to apply!