Fraunhofer-Gesellschaft
Masterarbeit: Lokalisierung von Sicherheitsneuronen in Text2Image-Modellen(m/w/x)
Full-timeOn-siteNot specified
Darmstadt
New Job?Nejo!
The AI Job Search Engine
FRFraunhofer-Gesellschaft
Master's Thesis: Localisation of Safety Neurones in Text2Image Models(m/w/x)
Darmstadt
Full-timeInternshipOn-site
Nejo AI Summary
Researching neuron localization in T2I models, adapting T2T approaches at an application-oriented research institute. Machine Learning knowledge (transformers) and Python (Pytorch) skills required. Independent work schedule management.
Requirements
- Knowledge in Machine Learning (training, inference, transformer architectures)
- Desirable knowledge in ML security
- Good Python skills (Pytorch)
- Scientific interest
- Interest in current research projects
Tasks
- Investigate safety neuron localizations in T2I models.
- Compare T2I neuron localizations with T2T models.
- Adapt T2T safety neuron localization approaches for T2I models.
- Implement adapted localization approaches for T2I models.
- Analyze neuron localization and significance results.
- Highlight architectural similarities and differences.
- Conceptualize model architectural differences for LLM security research.
- Motivate further research on T2I models.
- Conduct a comprehensive comparison of architectures and safety neurons.
- Verify results through targeted neuron deactivation.
- Compare results with randomly deactivated neurons.
- Research and implement novel machine learning approaches to enhance LLM security.
- Critically evaluate obtained results.
- Present research results.
- Prepare a project report as a master's thesis.
Education
- Currently in higher educationOR
- Bachelor's degreeOR
- Master's degree
Languages
- English – Business Fluent
Tools & Technologies
- Python
- Pytorch
- transformer architectures
Benefits
Flexible Working
- Independent work schedule management
Find the original job posting in its most current version here. Nejo automatically captured this job from the website of Fraunhofer-Gesellschaft and processed the information on Nejo with the help of AI for you. Despite careful analysis, some information may be incomplete or inaccurate. Please always verify all details in the original posting! Content and copyrights of the original posting belong to the advertising company.
Not a perfect match?
- Fraunhofer-Gesellschaft
Master's Thesis: Attacking Current Backdoor Detection Methods(m/w/x)
Full-timeOn-siteNot specifiedDarmstadt - Fraunhofer-Gesellschaft
Master's Thesis: Detection and Segmentation of Suggestive Clothing(m/w/x)
Full-timeInternshipOn-siteDarmstadt - Fraunhofer-Gesellschaft
Master's Thesis: Robustness of Authorship Verification against Adversarial Obfuscation(m/w/x)
Full-timeOn-siteNot specifiedDarmstadt - Fraunhofer-Gesellschaft
Masterarbeit: Aktuelle Backdoor Detektionsverfahren angreifen(m/w/x)
Full-timeInternshipOn-siteExperiencedDarmstadt
FRFraunhofer-Gesellschaft
Master's Thesis: Localisation of Safety Neurones in Text2Image Models(m/w/x)
Darmstadt
Full-timeInternshipOn-site
Nejo AI Summary
Researching neuron localization in T2I models, adapting T2T approaches at an application-oriented research institute. Machine Learning knowledge (transformers) and Python (Pytorch) skills required. Independent work schedule management.
Requirements
- Knowledge in Machine Learning (training, inference, transformer architectures)
- Desirable knowledge in ML security
- Good Python skills (Pytorch)
- Scientific interest
- Interest in current research projects
Tasks
- Investigate safety neuron localizations in T2I models.
- Compare T2I neuron localizations with T2T models.
- Adapt T2T safety neuron localization approaches for T2I models.
- Implement adapted localization approaches for T2I models.
- Analyze neuron localization and significance results.
- Highlight architectural similarities and differences.
- Conceptualize model architectural differences for LLM security research.
- Motivate further research on T2I models.
- Conduct a comprehensive comparison of architectures and safety neurons.
- Verify results through targeted neuron deactivation.
- Compare results with randomly deactivated neurons.
- Research and implement novel machine learning approaches to enhance LLM security.
- Critically evaluate obtained results.
- Present research results.
- Prepare a project report as a master's thesis.
Education
- Currently in higher educationOR
- Bachelor's degreeOR
- Master's degree
Languages
- English – Business Fluent
Tools & Technologies
- Python
- Pytorch
- transformer architectures
Benefits
Flexible Working
- Independent work schedule management
Find the original job posting in its most current version here. Nejo automatically captured this job from the website of Fraunhofer-Gesellschaft and processed the information on Nejo with the help of AI for you. Despite careful analysis, some information may be incomplete or inaccurate. Please always verify all details in the original posting! Content and copyrights of the original posting belong to the advertising company.
About the Company
Fraunhofer-Gesellschaft
Industry
Research
Description
Das Unternehmen ist eine der führenden Organisationen für anwendungsorientierte Forschung mit 76 Instituten in Deutschland.
Not a perfect match?
- Fraunhofer-Gesellschaft
Masterarbeit: Lokalisierung von Sicherheitsneuronen in Text2Image-Modellen(m/w/x)
Full-timeOn-siteNot specifiedDarmstadt - Fraunhofer-Gesellschaft
Master's Thesis: Attacking Current Backdoor Detection Methods(m/w/x)
Full-timeOn-siteNot specifiedDarmstadt - Fraunhofer-Gesellschaft
Master's Thesis: Detection and Segmentation of Suggestive Clothing(m/w/x)
Full-timeInternshipOn-siteDarmstadt - Fraunhofer-Gesellschaft
Master's Thesis: Robustness of Authorship Verification against Adversarial Obfuscation(m/w/x)
Full-timeOn-siteNot specifiedDarmstadt - Fraunhofer-Gesellschaft
Masterarbeit: Aktuelle Backdoor Detektionsverfahren angreifen(m/w/x)
Full-timeInternshipOn-siteExperiencedDarmstadt