New SELF method robustly fingerprints LLMs against IP theft

Researchers have developed a new method called SELF to protect the intellectual property of large language models (LLMs) against theft. This approach uses singular value and eigenvalue decomposition of LLM attention weights to create unique, transformation-invariant fingerprints. SELF is designed to be robust against various attacks, including quantization, pruning, and fine-tuning, and significantly reduces false positives compared to existing methods. AI

IMPACT This new fingerprinting technique could enhance the security and provenance tracking of LLMs, potentially deterring model theft and misuse.

RANK_REASON The cluster contains an academic paper detailing a new method for LLM fingerprinting. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hanxiu Zhang
• Hugging Face
• intellectual property
• large-language models
• ScienceCast
• SELF

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →