Differentially Private Neural Networks with Bounded Activation Function

Kijung JUNG; Hyukki LEE; Yon Dohn CHUNG

doi:10.1587/transinf.2021EDL8007

Differentially Private Neural Networks with Bounded Activation Function

Kijung JUNG, Hyukki LEE, Yon Dohn CHUNG

Full Text Views

0

Cite this

Summary :

Deep learning has shown outstanding performance in various fields, and it is increasingly deployed in privacy-critical domains. If sensitive data in the deep learning model are exposed, it can cause serious privacy threats. To protect individual privacy, we propose a novel activation function and stochastic gradient descent for applying differential privacy to deep learning. Through experiments, we show that the proposed method can effectively protect the privacy and the performance of proposed method is better than the previous approaches.

Publication: IEICE TRANSACTIONS on Information Vol.E104-D No.6 pp.905-908

Publication Date: 2021/06/01

Publicized: 2021/03/18

Online ISSN: 1745-1361

DOI: 10.1587/transinf.2021EDL8007

Type of Manuscript: LETTER

Category: Artificial Intelligence, Data Mining

Authors

Kijung JUNG
  Korea University
Hyukki LEE
  Korea University
Yon Dohn CHUNG
  Korea University

Keyword

deep learning, activation function, differential privacy

Cite this

Copy

Kijung JUNG, Hyukki LEE, Yon Dohn CHUNG, "Differentially Private Neural Networks with Bounded Activation Function" in IEICE TRANSACTIONS on Information, vol. E104-D, no. 6, pp. 905-908, June 2021, doi: 10.1587/transinf.2021EDL8007.
Abstract: Deep learning has shown outstanding performance in various fields, and it is increasingly deployed in privacy-critical domains. If sensitive data in the deep learning model are exposed, it can cause serious privacy threats. To protect individual privacy, we propose a novel activation function and stochastic gradient descent for applying differential privacy to deep learning. Through experiments, we show that the proposed method can effectively protect the privacy and the performance of proposed method is better than the previous approaches.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2021EDL8007/_p

Copy

@ARTICLE{e104-d_6_905,
author={Kijung JUNG, Hyukki LEE, Yon Dohn CHUNG, },
journal={IEICE TRANSACTIONS on Information},
title={Differentially Private Neural Networks with Bounded Activation Function},
year={2021},
volume={E104-D},
number={6},
pages={905-908},
abstract={Deep learning has shown outstanding performance in various fields, and it is increasingly deployed in privacy-critical domains. If sensitive data in the deep learning model are exposed, it can cause serious privacy threats. To protect individual privacy, we propose a novel activation function and stochastic gradient descent for applying differential privacy to deep learning. Through experiments, we show that the proposed method can effectively protect the privacy and the performance of proposed method is better than the previous approaches.},
keywords={},
doi={10.1587/transinf.2021EDL8007},
ISSN={1745-1361},
month={June},}

Copy

TY - JOUR
TI - Differentially Private Neural Networks with Bounded Activation Function
T2 - IEICE TRANSACTIONS on Information
SP - 905
EP - 908
AU - Kijung JUNG
AU - Hyukki LEE
AU - Yon Dohn CHUNG
PY - 2021
DO - 10.1587/transinf.2021EDL8007
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E104-D
IS - 6
JA - IEICE TRANSACTIONS on Information
Y1 - June 2021
AB - Deep learning has shown outstanding performance in various fields, and it is increasingly deployed in privacy-critical domains. If sensitive data in the deep learning model are exposed, it can cause serious privacy threats. To protect individual privacy, we propose a novel activation function and stochastic gradient descent for applying differential privacy to deep learning. Through experiments, we show that the proposed method can effectively protect the privacy and the performance of proposed method is better than the previous approaches.
ER -