Proving membership in LLM pretraining data via data watermarks
Authors:
- Johnny WeiJWJohnny Wei
- Ryan WangRWRyan Wang
- Robin JiaRJRobin Jia
Top Cited Papers
Open Access
- 1 January 2024
- conference paper
- Published by Association for Computational Linguistics (ACL)
- p. 13306-13320
Abstract
Johnny Wei, Ryan Wang, Robin Jia. Findings of the Association for Computational Linguistics ACL 2024. 2024.Keywords
This publication has 12 references indexed in Scilit:
- Not What You've Signed Up For: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection2023
- Reclaiming the Digital Commons: A Public Data Trust for Training Data2023
- Did You Train on My Dataset? Towards Public Dataset Protection with CleanLabel Backdoor WatermarkingACM SIGKDD Explorations Newsletter, 2023
- Membership Inference Attacks From First Principles2022
- Membership Inference Attacks on Machine Learning: A SurveyACM Computing Surveys, 2022
- BadNL: Backdoor Attacks against NLP Models with Semantic-preserving Improvements2021
- Model Cards for Model Reporting2019
- Membership Inference Attacks Against Machine Learning Models2017
- Deep Learning with Differential Privacy2016
- The MD5 Message-Digest Algorithm1992