Lorenz Hufe
Scholar

Lorenz Hufe

Google Scholar ID: BlI-rmsAAAAJ
Fraunhofer HHI
InterpretabilityComputer VisionNatural Language Processing
Homepage↗Google Scholar↗
Citations & Impact
All-time
Citations
15
 
H-index
3
 
i10-index
0
 
Publications
7
 
Co-authors
0
 
Contact
Emaillorenz.hufe@bliss.berlinGitHubOpen ↗LinkedInOpen ↗
Publications
7 items
Towards the AI Historian: Agentic Information Extraction from Primary Sources
2026
Cited
0
Towards Mechanistic Defenses Against Typographic Attacks in CLIP
2025
Cited
0
The TUB Sign Language Corpus Collection
2025
Cited
0
From What to How: Attributing CLIP's Latent Components Reveals Unexpected Semantic Reliance
2025
Cited
0
Prisma: An Open Source Toolkit for Mechanistic Interpretability in Vision and Video
2025
Cited
0
Steering CLIP's vision transformer with sparse autoencoders
2025
Cited
0
SCAM: A Real-World Typographic Robustness Evaluation for Multimodal Foundation Models
2025
Cited
0
Resume
Academic Achievements
  • Published 'Towards Mechanistic Defenses Against Typographic Attacks in CLIP', arXiv:2508.20570 (2025)
  • Contributed to 'The TUB Sign Language Corpus Collection', arXiv:2508.05374 (2025)
  • Co-authored 'From What to How: Attributing CLIP's Latent Components Reveals Unexpected Semantic Reliance', arXiv:2505.20229 (2025)
  • Core contributor to the Prisma open-source toolkit; presented orally at the Mechanistic Interpretability for Vision Workshop @ CVPR 2025
  • Co-authored 'Steering CLIP's Vision Transformer with Sparse Autoencoders', presented at CVPR 2025 Workshop
  • Co-authored 'SCAM: A Real-World Typographic Robustness Evaluation for Multimodal Foundation Models', presented at CVPR 2025 Workshop
  • Published 'Experimental Machine Translation of the Swiss German Sign Language via 3D Augmentation of Body Keypoints' at WMT 2022