Deterministic and Statistical Strategies to Protect ANNs against Fault Injection Attacks

T.C. Köylü; Cezar Reinbrecht; S. Hamdioui; M. Taouil

doi:10.1109/PST52912.2021.9647763

Deterministic and Statistical Strategies to Protect ANNs against Fault Injection Attacks

T.C. Köylü^*, Cezar Reinbrecht, S. Hamdioui, M. Taouil

^*Corresponding author for this work

Research output: Chapter in Book/Conference proceedings/Edited volume › Conference contribution › Scientific › peer-review

1 Citation (Scopus)

24 Downloads (Pure)

Abstract

Artificial neural networks are currently used for many tasks, including safety critical ones such as automated driving. Hence, it is very important to protect them against faults and fault attacks. In this work, we propose two fault injection attack detection mechanisms: one based on using output labels for a reference input, and the other on the activations of neurons. First, we calibrate our detectors during normal conditions. Thereafter, we verify them to maximize fault detection performance. To prove the effectiveness of our solution, we consider highly employed neural networks (AlexNet, GoogleNet, and VGG) with their associated dataset ImageNet. Our results show that for both detectors we are able to obtain a high rate of coverage against faults, typically above 96%. Moreover, the hardware and software implementations of our detector indicate an extremely low area and time overhead.

Original language	English
Title of host publication	2021 18th International Conference on Privacy, Security and Trust (PST)
Place of Publication	Piscataway
Publisher	IEEE
Pages	1-10
Number of pages	10
ISBN (Electronic)	978-1-6654-0184-5
ISBN (Print)	978-1-6654-0185-2
DOIs	https://doi.org/10.1109/PST52912.2021.9647763
Publication status	Published - 2021
Event	18th Annual International Conference on Privacy, Security and Trust (PST2021) - Virtual at Auckland, New Zealand Duration: 13 Dec 2021 → 15 Dec 2021 Conference number: 18

Publication series

Name	2021 18th International Conference on Privacy, Security and Trust, PST 2021

Conference

Conference	18th Annual International Conference on Privacy, Security and Trust (PST2021)
Abbreviated title	PST2021
Country/Territory	New Zealand
City	Virtual at Auckland
Period	13/12/21 → 15/12/21

Keywords

Fault Injection
Countermeasures
Artificial neural networks
Machine learning

Access to Document

10.1109/PST52912.2021.9647763

paperAccepted author manuscript, 547 KB

Cite this

Köylü, T. C., Reinbrecht, C., Hamdioui, S., & Taouil, M. (2021). Deterministic and Statistical Strategies to Protect ANNs against Fault Injection Attacks. In 2021 18th International Conference on Privacy, Security and Trust (PST) (pp. 1-10). (2021 18th International Conference on Privacy, Security and Trust, PST 2021). IEEE. https://doi.org/10.1109/PST52912.2021.9647763

@inproceedings{341ae459f1a44346aff945c7ed14c446,

title = "Deterministic and Statistical Strategies to Protect ANNs against Fault Injection Attacks",

abstract = "Artificial neural networks are currently used for many tasks, including safety critical ones such as automated driving. Hence, it is very important to protect them against faults and fault attacks. In this work, we propose two fault injection attack detection mechanisms: one based on using output labels for a reference input, and the other on the activations of neurons. First, we calibrate our detectors during normal conditions. Thereafter, we verify them to maximize fault detection performance. To prove the effectiveness of our solution, we consider highly employed neural networks (AlexNet, GoogleNet, and VGG) with their associated dataset ImageNet. Our results show that for both detectors we are able to obtain a high rate of coverage against faults, typically above 96%. Moreover, the hardware and software implementations of our detector indicate an extremely low area and time overhead.",

keywords = "Fault Injection, Countermeasures, Artificial neural networks, Machine learning",

author = "T.C. K{\"o}yl{\"u} and Cezar Reinbrecht and S. Hamdioui and M. Taouil",

year = "2021",

doi = "10.1109/PST52912.2021.9647763",

language = "English",

isbn = "978-1-6654-0185-2",

series = "2021 18th International Conference on Privacy, Security and Trust, PST 2021",

publisher = "IEEE",

pages = "1--10",

booktitle = "2021 18th International Conference on Privacy, Security and Trust (PST)",

address = "United States",

note = "18th Annual International Conference on Privacy, Security and Trust (PST2021), PST2021 ; Conference date: 13-12-2021 Through 15-12-2021",

}

Köylü, TC, Reinbrecht, C, Hamdioui, S & Taouil, M 2021, Deterministic and Statistical Strategies to Protect ANNs against Fault Injection Attacks. in 2021 18th International Conference on Privacy, Security and Trust (PST). 2021 18th International Conference on Privacy, Security and Trust, PST 2021, IEEE, Piscataway, pp. 1-10, 18th Annual International Conference on Privacy, Security and Trust (PST2021), Virtual at Auckland, New Zealand, 13/12/21. https://doi.org/10.1109/PST52912.2021.9647763

Deterministic and Statistical Strategies to Protect ANNs against Fault Injection Attacks. / Köylü, T.C.; Reinbrecht, Cezar; Hamdioui, S. et al.
2021 18th International Conference on Privacy, Security and Trust (PST). Piscataway: IEEE, 2021. p. 1-10 (2021 18th International Conference on Privacy, Security and Trust, PST 2021).

Research output: Chapter in Book/Conference proceedings/Edited volume › Conference contribution › Scientific › peer-review

TY - GEN

T1 - Deterministic and Statistical Strategies to Protect ANNs against Fault Injection Attacks

AU - Köylü, T.C.

AU - Reinbrecht, Cezar

AU - Hamdioui, S.

AU - Taouil, M.

N1 - Conference code: 18

PY - 2021

Y1 - 2021

N2 - Artificial neural networks are currently used for many tasks, including safety critical ones such as automated driving. Hence, it is very important to protect them against faults and fault attacks. In this work, we propose two fault injection attack detection mechanisms: one based on using output labels for a reference input, and the other on the activations of neurons. First, we calibrate our detectors during normal conditions. Thereafter, we verify them to maximize fault detection performance. To prove the effectiveness of our solution, we consider highly employed neural networks (AlexNet, GoogleNet, and VGG) with their associated dataset ImageNet. Our results show that for both detectors we are able to obtain a high rate of coverage against faults, typically above 96%. Moreover, the hardware and software implementations of our detector indicate an extremely low area and time overhead.

AB - Artificial neural networks are currently used for many tasks, including safety critical ones such as automated driving. Hence, it is very important to protect them against faults and fault attacks. In this work, we propose two fault injection attack detection mechanisms: one based on using output labels for a reference input, and the other on the activations of neurons. First, we calibrate our detectors during normal conditions. Thereafter, we verify them to maximize fault detection performance. To prove the effectiveness of our solution, we consider highly employed neural networks (AlexNet, GoogleNet, and VGG) with their associated dataset ImageNet. Our results show that for both detectors we are able to obtain a high rate of coverage against faults, typically above 96%. Moreover, the hardware and software implementations of our detector indicate an extremely low area and time overhead.

KW - Fault Injection

KW - Countermeasures

KW - Artificial neural networks

KW - Machine learning

UR - http://www.scopus.com/inward/record.url?scp=85124081637&partnerID=8YFLogxK

U2 - 10.1109/PST52912.2021.9647763

DO - 10.1109/PST52912.2021.9647763

M3 - Conference contribution

SN - 978-1-6654-0185-2

T3 - 2021 18th International Conference on Privacy, Security and Trust, PST 2021

SP - 1

EP - 10

BT - 2021 18th International Conference on Privacy, Security and Trust (PST)

PB - IEEE

CY - Piscataway

T2 - 18th Annual International Conference on Privacy, Security and Trust (PST2021)

Y2 - 13 December 2021 through 15 December 2021

ER -

Deterministic and Statistical Strategies to Protect ANNs against Fault Injection Attacks

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this