Investigating How Pre-training Data Leakage Affects Models’ Reproduction and Detection Capabilities (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies do not examine how leaked instances in training datasets influence LLMs’ output and detection capabilities. |
| Approach: | They conduct an experimental survey to examine the relationship between data leakage in training datasets and its effects on the generation and detection by Large Language Models (LLMs). |
| Outcome: | The results show that enhancing leakage detection through few-shot learning can help mitigate the impact of the leakage rate in the training data on detection performance. |
Similar Papers
Does Data Contamination Detection Work (Well) for LLMs? A Survey and Evaluation on Detection Assumptions (2025.findings-naacl)
Copied to clipboard
| Challenge: | Large language models have demonstrated great performance across various benchmarks, but data contamination is a concern in their evaluation. |
| Approach: | They analyze 50 papers on data contamination detection and test three of them as case studies to identify the possibility of data contamination. |
| Outcome: | The proposed methods can detect membership inference attacks on instance-level data, and can perform similar to random guessing on LLM pretraining datasets. |
Are Large Pre-Trained Language Models Leaking Your Personal Information? (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Pre-trained language models (PLMs) are prone to leaking personal information due to memorization, but the risk of specific personal information being extracted by attackers is low. |
| Approach: | They analyze whether large pre-trained language models are prone to leaking personal information due to memorization. |
| Outcome: | The proposed model is weak at association, so the risk of specific personal information being extracted by attackers is low. |
Revisiting the Effects of Leakage on Dependency Parsing (2022.findings-acl)
Copied to clipboard
| Challenge: | Recent work shows that treebank size and linguistic variation are important factors that explain the variation in dependency parsing performance. |
| Approach: | They propose a measure of leakage that explains and correlates with observed performance variation. |
| Outcome: | The proposed measure explains and correlates with observed performance variation. |
Deciphering the Impact of Pretraining Data on Large Language Models through Machine Unlearning (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have suggested that the composition of the pretraining corpus exerts a significant impact upon the performance of LLMs. |
| Approach: | They analyze the impact of 48 datasets from 5 major categories of pretraining data of Large Language Models and measure their impacts on LLMs using benchmarks about nine major categories. |
| Outcome: | The proposed analysis provides insights into the organization of data to support more efficient pretraining of Large Language Models. |
Leak, Cheat, Repeat: Data Contamination and Evaluation Malpractices in Closed-Source LLMs (2024.eacl-long)
Copied to clipboard
| Challenge: | Lack of access to model details has raised concerns about data contamination among researchers. |
| Approach: | They conduct the first systematic analysis of work using OpenAI’s GPT-3.5 and GPT-4, the most prominently used LLMs today, in the context of data contamination. |
| Outcome: | The proposed models have been exposed to 4.7M samples from 263 benchmarks during the first year after their release. |
SrDetection: A Self-Referential Framework for Data Leakage Detection in Code Large Language Models (2026.findings-acl)
Copied to clipboard
Shuaimin Li, Liyang Fan, Zeyang li, Zhuoyue Wan, Yufang Lin, Shiwen Ni, Feiteng Fang, Hamid Alinejad-Rokny, Yuanfeng Song, Kun Jing, Chen Jason Zhang, Min Yang
| Challenge: | Existing methods for evaluating code large language models assume access to proprietary training corpora or use external reference sets with manually tuned, non-generalizable thresholds. |
| Approach: | They propose a framework for self-referential leakage detection for gray-box and black-box settings. |
| Outcome: | The proposed framework improves average F1 by 21.52 points in the gray-box setting and 14.46 points in black-box settings over strong baselines. |
On Leakage of Code Generation Evaluation Datasets (2024.findings-emnlp)
Copied to clipboard
Alexandre Matton, Tom Sherborne, Dennis Aumiller, Elena Tommasone, Milad Alizadeh, Jingyi He, Raymond Ma, Maxime Voisin, Ellen Gilsenan-McMahon, Matthias Gallé
| Challenge: | In this paper, we discuss contamination by code generation test sets in large language models. |
| Approach: | They propose to use Python to test code generation test sets for contamination . they find that code generation is an important skill for large language models to master . |
| Outcome: | The proposed benchmarks are uncontaminated and provide a new insight into code generation. |
Adaptation Odyssey in LLMs: Why Does Additional Pretraining Sometimes Fail to Improve? (2024.emnlp-main)
Copied to clipboard
| Challenge: | In the last decade, the generalization and adaptation abilities of deep learning models were evaluated on fixed training and test distributions. |
| Approach: | They propose to train large language models on unlabeled text corpora and train them online. |
| Outcome: | The proposed model training on a text domain could degrade its perplexity on the test portion of the same domain. |
User Inference Attacks on Large Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | a large amount of data written by humans is used to train and fine-tune large language models. |
| Approach: | They propose to infer if a user's data was used to train an LLM by using example-level differential privacy. |
| Outcome: | The proposed attacks are easy to employ and only require black-box access to an LLM and a few samples from the user. |
SecureSQL: Evaluating Data Leakage of Large Language Models as Natural Language Interfaces to Databases (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies on the vulnerability of large language models to SQL injection have been limited. |
| Approach: | They propose to evaluate the potential of language models to leak sensitive data when generating SQL queries. |
| Outcome: | The proposed model with the best performance has an accuracy of 61.7%, compared to humans who achieve 94% accuracy. |