Student Assistant – LLM Security
Fraunhofer-Gesellschaft
part-time
Posted on:
Location Type: Hybrid
Location: Darmstadt • Germany
Visit company websiteExplore more
Job Level
About the role
- Development of methods to enhance the security of LLMs.
- A major focus is on defending against common attacks such as backdoor attacks and associated jailbreaks.
- Systematic evaluation of the robustness of LLMs against established and novel attack scenarios, regarding common and novel attacks such as backdoor attacks, divergence attacks, or membership inference attacks, etc.
- Investigation, application, and evaluation of post-hoc explainability methods to analyze model behavior, with the aim of gaining a deeper understanding of the underlying attack mechanisms as well as the effectiveness and limitations of corresponding defense strategies.
Requirements
- Very good Python skills, which will be tested in the interview.
- Basic understanding of AI/ML mainly in relation to transformers
- Experience with Pytorch and transformer architectures is desirable
- Good English skills
- Enrolled student (Computer Science, Data Science, Mathematics or similar)
- Structured, independent working style and ability to work in a team
- Strong analytical thinking
Benefits
- Flexible working hours that can be easily combined with your studies
- An inspiring work environment with state-of-the-art infrastructure
- The opportunity to gain practical experience and make valuable contacts in research
- Opportunity to collaborate on scientific papers and write a later bachelor's or master's thesis.
Applicant Tracking System Keywords
Tip: use these terms in your resume and cover letter to boost ATS matches.
Hard Skills & Tools
PythonAIMachine LearningPytorchtransformer architecturespost-hoc explainability methodsrobustness evaluationattack scenariosbackdoor attacksmembership inference attacks
Soft Skills
analytical thinkingstructured working styleindependent workingteamworkcommunication skills