Training Data Extraction

Training data extraction focuses on identifying and retrieving fragments of the datasets used to train large language models (LLMs), revealing potential vulnerabilities in data privacy and security. Current research explores various techniques, including prompting strategies (e.g., using special characters or carefully crafted instructions) and refined ranking methods to improve the accuracy and efficiency of data extraction. This research is crucial for assessing the security risks associated with LLMs and informing the development of more robust privacy-preserving training and deployment methods.

Papers

September 18, 2024

Extracting Memorized Training Data via Decomposition
Ellen Su, Anu Vellore, Amy Chang, Raffaele Mura, Blaine Nelson, Paul Kassianik, Amin Karbasi
Training Data Microbial Decomposition Training Data Privacy Source Training Training Data Extraction

May 9, 2024

Special Characters Attack: Toward Scalable Training Data Extraction From Large Language Models
Yang Bai, Ge Pei, Jindong Gu, Yong Yang, Xingjun Ma
Large Language Model Training Corpus Scale Pre Trained Language Model Training Data Leakage Training Data Extraction

July 10, 2023

Ethicist: Targeted Training Data Extraction Through Loss Smoothed Soft Prompting and Calibrated Confidence Estimation
Zhexin Zhang, Jiaxin Wen, Minlie Huang
Pre Trained Language Model Large Pre Trained Language Model Intercultural Ethic Calibrated Confidence Suffix Prediction Training Data Extraction

May 25, 2023

Training Data Extraction From Pre-trained Language Models: A Survey
Shotaro Ishihara
Natural Language Processing Timely Survey Training Data Pre Trained Language Model Extraction Attack Training Data Extraction

February 9, 2023

Bag of Tricks for Training Data Extraction from Language Models
Weichen Yu, Tianyu Pang, Qian Liu, Chao Du, Bingyi Kang, Yan Huang, Min Lin, Shuicheng Yan
Language Model Text Generation Bag Prototype Unconventional Rabbit Hat Trick High Quality Training Datasets Extraction Method Training Data Extraction

Training Data Extraction

Papers

Extracting Memorized Training Data via Decomposition

Special Characters Attack: Toward Scalable Training Data Extraction From Large Language Models

Ethicist: Targeted Training Data Extraction Through Loss Smoothed Soft Prompting and Calibrated Confidence Estimation

Training Data Extraction From Pre-trained Language Models: A Survey

Bag of Tricks for Training Data Extraction from Language Models