Researchers have introduced a new method for entity-level membership inference in large language models (LLMs). This approach aims to determine if information about a specific real-world entity, rather than just individual data samples, was included in an LLM's training data. The proposed strategies involve using limited entity clues to construct prompts, elicit responses, and then infer membership based on semantic features within the generated text. Experiments demonstrated that these methods can achieve high accuracy, significantly outperforming adapted baseline approaches. AI
IMPACT This research could lead to better privacy auditing for LLMs, helping to identify if sensitive entity-level information is being memorized during training.
RANK_REASON The item is a research paper detailing a new method for membership inference in LLMs. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
- entity-level membership inference
- Have You Ever Seen Them? Entity-level Membership Inference through Interrogating Large Language Models
- Hugging Face
- large-language models
- Membership inference attack
- person entities
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
