SAP: Privacy-Preserving Fine-Tuning on Language Models with Split-and-Privatize Framework

Shen, Xicong; Liu, Yang; Liu, Yi; Wang, Peiran; Liu, Huiqi; Hong, Jue; Duan, Bing; Huang, Zirui; Mao, Yunlong; Wu, Ye; Zhong, Sheng

doi:10.24963/ijcai.2025/57

SAP: Privacy-Preserving Fine-Tuning on Language Models with Split-and-Privatize Framework

Xicong Shen, Yang Liu, Yi Liu, Peiran Wang, Huiqi Liu, Jue Hong, Bing Duan, Zirui Huang, Yunlong Mao, Ye Wu, Sheng Zhong

Download BibTeX Cite

Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence. Main Track. Pages 502-510.

https://doi.org/10.24963/ijcai.2025/57

Keywords

Multidisciplinary Topics and Applications: MTA: Security and privacy
AI Ethics, Trust, Fairness: ETF: Trustworthy AI
Machine Learning: ML: Federated learning
Machine Learning: ML: Trustworthy machine learning

Abstract

Pre-trained Language Models (PLM) have enabled a cost-effective approach to handling various downstream applications via Parameter-Efficient-Fine-Tuning (PEFT) techniques. In this context, service providers have introduced a popular fine-tuning-based product service known as Model-as-a-Service (MaaS). This service offers users access to extensive PLMs and training resources. With MaaS, users can fine-tune, deploy, and utilize their customized models seamlessly, leveraging a one-stop platform that allows them to work with their private datasets efficiently. However, this service paradigm has recently been exposed to the possibility of leaking user private data. To this end, we identify the data privacy leakage risks in MaaS-based PEFT and propose a Split-and-Privatize (SAP) framework, mitigating the privacy leakage by integrating split learning and differential privacy into MaaS PEFT. Furthermore, we propose Contributing-Token-Identification (CTI), a novel method to balance model utility degradation and privacy leakage. As a result, the proposed framework is comprehensively evaluated, demonstrating a 65% improvement in empirical privacy with only a 1% degradation in model performance on the Stanford Sentiment Treebank dataset, outperforming existing state-of-the-art baselines.

Xicong Shen, Yang Liu, Yi Liu, Peiran Wang, Huiqi Liu, Jue Hong, Bing Duan, Zirui Huang, Yunlong Mao, Ye Wu, and Sheng Zhong. 2025. SAP: Privacy-Preserving Fine-Tuning on Language Models with Split-and-Privatize Framework. In Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25. IJCAI, 502–510. https://doi.org/10.24963/ijcai.2025/57

@inproceedings{ijcai2025p57,
  title     = {SAP: Privacy-Preserving Fine-Tuning on Language Models with Split-and-Privatize Framework},
  author    = {Shen, Xicong and Liu, Yang and Liu, Yi and Wang, Peiran and Liu, Huiqi and Hong, Jue and Duan, Bing and Huang, Zirui and Mao, Yunlong and Wu, Ye and Zhong, Sheng},
  booktitle = {Proceedings of the Thirty-Fourth International Joint Conference on
               Artificial Intelligence, {IJCAI-25}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {James Kwok},
  pages     = {502--510},
  year      = {2025},
  month     = {8},
  note      = {Main Track},
  doi       = {10.24963/ijcai.2025/57},
  url       = {https://doi.org/10.24963/ijcai.2025/57},
}

TY  - CPAPER
TI  - SAP: Privacy-Preserving Fine-Tuning on Language Models with Split-and-Privatize Framework
AU  - Shen, Xicong
AU  - Liu, Yang
AU  - Liu, Yi
AU  - Wang, Peiran
AU  - Liu, Huiqi
AU  - Hong, Jue
AU  - Duan, Bing
AU  - Huang, Zirui
AU  - Mao, Yunlong
AU  - Wu, Ye
AU  - Zhong, Sheng
T2  - Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25
PB  - International Joint Conferences on Artificial Intelligence Organization
PY  - 2025
SP  - 502
EP  - 510
DO  - 10.24963/ijcai.2025/57
UR  - https://doi.org/10.24963/ijcai.2025/57
ER  -