Privacy-Preserving Student Learning with Differentially Private Data-Free Distillation

被引:2
|
作者
Liu, Bochao [1 ,2 ]
Lu, Jianghu [1 ,2 ]
Wang, Pengju [1 ,2 ]
Zhang, Junjie [3 ]
Zeng, Dan [3 ]
Qian, Zhenxing [4 ]
Ge, Shiming [1 ,2 ]
机构
[1] Chinese Acad Sci, Inst Informat Engn, Beijing 100095, Peoples R China
[2] Univ Chinese Acad Sci, Sch Cyber Secur, Beijing 100049, Peoples R China
[3] Shanghai Univ, Sch Commun & Informat Engn, Shanghai 200444, Peoples R China
[4] Fudan Univ, Sch Comp Sci, Shanghai 200433, Peoples R China
来源
2022 IEEE 24TH INTERNATIONAL WORKSHOP ON MULTIMEDIA SIGNAL PROCESSING (MMSP) | 2022年
基金
北京市自然科学基金;
关键词
differential privacy; teacher-student learning; knowledge distillation;
D O I
10.1109/MMSP55362.2022.9950001
中图分类号
TP31 [计算机软件];
学科分类号
081202 ; 0835 ;
摘要
Deep learning models can achieve high inference accuracy by extracting rich knowledge from massive well-annotated data, but may pose the risk of data privacy leakage in practical deployment. In this paper, we present an effective teacher-student learning approach to train privacy-preserving deep learning models via differentially private data-free distillation. The main idea is generating synthetic data to learn a student that can mimic the ability of a teacher well-trained on private data. In the approach, a generator is first pretrained in a data-free manner by incorporating the teacher as a fixed discriminator. With the generator, massive synthetic data can be generated for model training without exposing data privacy. Then, the synthetic data is fed into the teacher to generate private labels. Towards this end, we propose a label differential privacy algorithm termed selective randomized response to protect the label information. Finally, a student is trained on the synthetic data with the supervision of private labels. In this way, both data privacy and label privacy are well protected in a unified framework, leading to privacy-preserving models. Extensive experiments and analysis clearly demonstrate the effectiveness of our approach.
引用
收藏
页数:6
相关论文
共 50 条
  • [1] Data-Free Knowledge Distillation for Privacy-Preserving Efficient UAV Networks
    Yu, Guyang
    2022 6TH INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION SCIENCES (ICRAS 2022), 2022, : 52 - 56
  • [2] Privacy-Preserving Federated Learning with Differentially Private Hyperdimensional Computing
    Piran, Fardin Jalil
    Chen, Zhiling
    Imani, Mohsen
    Imani, Farhad
    COMPUTERS & ELECTRICAL ENGINEERING, 2025, 123
  • [3] Learning Privacy-Preserving Student Networks via Discriminative-Generative Distillation
    Ge, Shiming
    Liu, Bochao
    Wang, Pengju
    Li, Yong
    Zeng, Dan
    IEEE TRANSACTIONS ON IMAGE PROCESSING, 2023, 32 : 116 - 127
  • [4] Differentially Private Normalizing Flows for Privacy-Preserving Density Estimation
    Waites, Chris
    Cummings, Rachel
    AIES '21: PROCEEDINGS OF THE 2021 AAAI/ACM CONFERENCE ON AI, ETHICS, AND SOCIETY, 2021, : 1000 - 1009
  • [5] Privacy preserving classification over differentially private data
    Zorarpaci, Ezgi
    Ozel, Selma Ayse
    WILEY INTERDISCIPLINARY REVIEWS-DATA MINING AND KNOWLEDGE DISCOVERY, 2021, 11 (03)
  • [6] Differentially Private Sparse Mapping for Privacy-Preserving Cross Domain Recommendation
    Liu, Weiming
    Zheng, Xiaolin
    Chen, Chaochao
    Hu, Mengling
    Liao, Xinting
    Wang, Fan
    Tan, Yanchao
    Meng, Dan
    Wang, Jun
    PROCEEDINGS OF THE 31ST ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA, MM 2023, 2023, : 6243 - 6252
  • [7] Privacy-Preserving Utility Verification of the Data Published by Non-Interactive Differentially Private Mechanisms
    Hua, Jingyu
    Tang, An
    Fang, Yixin
    Shen, Zhenyu
    Zhong, Sheng
    IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, 2016, 11 (10) : 2298 - 2311
  • [8] Memory efficient data-free distillation for continual learning
    Li, Xiaorong
    Wang, Shipeng
    Sun, Jian
    Xu, Zongben
    PATTERN RECOGNITION, 2023, 144
  • [9] Privacy-Preserving Fog Aggregation of Smart Grid Data Using Dynamic Differentially-Private Data Perturbation
    Kserawi, Fawaz
    Al-Marri, Saeed
    Malluhi, Qutaibah
    IEEE ACCESS, 2022, 10 : 43159 - 43174
  • [10] Differentially Privacy-preserving Social IoT
    Zhang, Linjie
    Zhu, Xiaoyan
    Han, Xuexue
    Ma, Jianfeng
    2019 11TH INTERNATIONAL CONFERENCE ON WIRELESS COMMUNICATIONS AND SIGNAL PROCESSING (WCSP), 2019,