SIT: Stochastic Input Transformation to Defend Against Adversarial Attacks on Deep Neural Networks

Amira Guesmi; Ihsen Alouani; Mouna Baklouti; Tarek Frikha; Mohamed Abid

doi:10.1109/MDAT.2021.3077542

SIT: Stochastic Input Transformation to Defend Against Adversarial Attacks on Deep Neural Networks

Amira Guesmi, Ihsen Alouani, Mouna Baklouti, Tarek Frikha, Mohamed Abid

Research output: Contribution to journal › Article › peer-review

10 Citations (Scopus)

Abstract

To better combat the impact of adversarial samples on deep neural networks, a model-agnostic stochastic input transformation (SIT) preprocessing technique is proposed in this article. The inputs are transformed into a new domain to minimize the impact of the adversarial perturbations.

Original language	English
Pages (from-to)	63-72
Number of pages	10
Journal	IEEE Design and Test
Volume	39
Issue number	3
Early online date	04 May 2021
DOIs	https://doi.org/10.1109/MDAT.2021.3077542
Publication status	Published - 01 Jun 2022
Externally published	Yes

Keywords

Adversarial Attacks
Convolutional Neural Networks
Machine Learning
Security

Access to Document

10.1109/MDAT.2021.3077542Licence: Unspecified

Cite this

@article{9df0762d83e64f4894127a83426c2999,

title = "SIT: Stochastic Input Transformation to Defend Against Adversarial Attacks on Deep Neural Networks",

abstract = "To better combat the impact of adversarial samples on deep neural networks, a model-agnostic stochastic input transformation (SIT) preprocessing technique is proposed in this article. The inputs are transformed into a new domain to minimize the impact of the adversarial perturbations.",

keywords = "Adversarial Attacks, Convolutional Neural Networks, Machine Learning, Security",

author = "Amira Guesmi and Ihsen Alouani and Mouna Baklouti and Tarek Frikha and Mohamed Abid",

year = "2022",

month = jun,

day = "1",

doi = "10.1109/MDAT.2021.3077542",

language = "English",

volume = "39",

pages = "63--72",

journal = "IEEE Design and Test",

issn = "2168-2364",

number = "3",

}

TY - JOUR

T1 - SIT: Stochastic Input Transformation to Defend Against Adversarial Attacks on Deep Neural Networks

AU - Guesmi, Amira

AU - Alouani, Ihsen

AU - Baklouti, Mouna

AU - Frikha, Tarek

AU - Abid, Mohamed

PY - 2022/6/1

Y1 - 2022/6/1

N2 - To better combat the impact of adversarial samples on deep neural networks, a model-agnostic stochastic input transformation (SIT) preprocessing technique is proposed in this article. The inputs are transformed into a new domain to minimize the impact of the adversarial perturbations.

AB - To better combat the impact of adversarial samples on deep neural networks, a model-agnostic stochastic input transformation (SIT) preprocessing technique is proposed in this article. The inputs are transformed into a new domain to minimize the impact of the adversarial perturbations.

KW - Adversarial Attacks

KW - Convolutional Neural Networks

KW - Machine Learning

KW - Security

U2 - 10.1109/MDAT.2021.3077542

DO - 10.1109/MDAT.2021.3077542

M3 - Article

SN - 2168-2364

VL - 39

SP - 63

EP - 72

JO - IEEE Design and Test

JF - IEEE Design and Test

IS - 3

ER -

SIT: Stochastic Input Transformation to Defend Against Adversarial Attacks on Deep Neural Networks

Abstract

Keywords

Access to Document

Fingerprint

Cite this