Saved in:
Bibliographic Details
Main Authors: Wang, Jianwei, Shi, Chengming, Yang, Junyao, Li, Haoran, Ma, Qianli, Zhuang, Huiping, Chen, Cen, Zeng, Ziqian
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.18517
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909761918730240
author Wang, Jianwei
Shi, Chengming
Yang, Junyao
Li, Haoran
Ma, Qianli
Zhuang, Huiping
Chen, Cen
Zeng, Ziqian
author_facet Wang, Jianwei
Shi, Chengming
Yang, Junyao
Li, Haoran
Ma, Qianli
Zhuang, Huiping
Chen, Cen
Zeng, Ziqian
contents The success of large language models (LLMs) has attracted many individuals to fine-tune them for domain-specific tasks by uploading their data. However, in sensitive areas like healthcare and finance, privacy concerns often arise. One promising solution is to generate synthetic data with Differential Privacy (DP) guarantees to replace private data. However, these synthetic data contain significant flawed data, which are considered as noise. Existing solutions typically rely on naive filtering by comparing ROUGE-L scores or embedding similarities, which are ineffective in addressing the noise. To address this issue, we propose \textit{RewardDS}, a novel privacy-preserving framework that fine-tunes a reward proxy model and uses reward signals to guide the synthetic data generation. Our \textit{RewardDS} introduces two key modules, Reward Guided Filtering and Self-Optimizing Refinement, to both filter and refine the synthetic data, effectively mitigating the noise. Extensive experiments across medical, financial, and code generation domains demonstrate the effectiveness of our method.
format Preprint
id arxiv_https___arxiv_org_abs_2502_18517
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle RewardDS: Privacy-Preserving Fine-Tuning for Large Language Models via Reward Driven Data Synthesis
Wang, Jianwei
Shi, Chengming
Yang, Junyao
Li, Haoran
Ma, Qianli
Zhuang, Huiping
Chen, Cen
Zeng, Ziqian
Cryptography and Security
Artificial Intelligence
The success of large language models (LLMs) has attracted many individuals to fine-tune them for domain-specific tasks by uploading their data. However, in sensitive areas like healthcare and finance, privacy concerns often arise. One promising solution is to generate synthetic data with Differential Privacy (DP) guarantees to replace private data. However, these synthetic data contain significant flawed data, which are considered as noise. Existing solutions typically rely on naive filtering by comparing ROUGE-L scores or embedding similarities, which are ineffective in addressing the noise. To address this issue, we propose \textit{RewardDS}, a novel privacy-preserving framework that fine-tunes a reward proxy model and uses reward signals to guide the synthetic data generation. Our \textit{RewardDS} introduces two key modules, Reward Guided Filtering and Self-Optimizing Refinement, to both filter and refine the synthetic data, effectively mitigating the noise. Extensive experiments across medical, financial, and code generation domains demonstrate the effectiveness of our method.
title RewardDS: Privacy-Preserving Fine-Tuning for Large Language Models via Reward Driven Data Synthesis
topic Cryptography and Security
Artificial Intelligence
url https://arxiv.org/abs/2502.18517