Are self-driving cars secure? Evasion attacks against deep neural networks for steering angle prediction

Alesia Chernikova; Alina Oprea; Cristina Nita-Rotaru; Baekgyu Kim

doi:10.1109/SPW.2019.00033

Are self-driving cars secure? Evasion attacks against deep neural networks for steering angle prediction

Alesia Chernikova
, Alina Oprea
, Cristina Nita-Rotaru
, Baekgyu Kim

Department of Electrical Engineering and Computer Science

Northeastern University

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

60 Scopus citations

Abstract

Deep Neural Networks (DNNs) have tremendous potential in advancing the vision for self-driving cars. However, the security of DNN models in this context leads to major safety implications and needs to be better understood. We consider the case study of steering angle prediction from camera images, using the dataset from the 2014 Udacity challenge. We demonstrate for the first time adversarial testing-time attacks for this application for both classification and regression settings. We show that minor modifications to the camera image (an L-2 distance of 0.82 for one of the considered models) result in mis-classification of an image to any class of attacker's choice. Furthermore, our regression attack results in a significant increase in Mean Square Error (MSE) - by a factor of 69 in the worst case.

Original language	English
Title of host publication	Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	132-137
Number of pages	6
ISBN (Electronic)	9781728135083
DOIs	https://doi.org/10.1109/SPW.2019.00033
State	Published - May 2019
Event	2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019 - San Francisco, United States Duration: 23 May 2019 → …

Publication series

Name	Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019

Conference

Conference	2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019
Country/Territory	United States
City	San Francisco
Period	23/05/19 → …

Bibliographical note

Publisher Copyright:
© 2019 IEEE.

Keywords

Adversarial machine learning
Deep neural networks
Self driving cars

Access to Document

10.1109/SPW.2019.00033

Cite this

Chernikova, A., Oprea, A., Nita-Rotaru, C., & Kim, B. (2019). Are self-driving cars secure? Evasion attacks against deep neural networks for steering angle prediction. In Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019 (pp. 132-137). Article 8844593 (Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/SPW.2019.00033

Chernikova, Alesia ; Oprea, Alina ; Nita-Rotaru, Cristina et al. / Are self-driving cars secure? Evasion attacks against deep neural networks for steering angle prediction. Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019. Institute of Electrical and Electronics Engineers Inc., 2019. pp. 132-137 (Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019).

@inproceedings{42a06294b3964f92981ec56c263851bc,

title = "Are self-driving cars secure? Evasion attacks against deep neural networks for steering angle prediction",

abstract = "Deep Neural Networks (DNNs) have tremendous potential in advancing the vision for self-driving cars. However, the security of DNN models in this context leads to major safety implications and needs to be better understood. We consider the case study of steering angle prediction from camera images, using the dataset from the 2014 Udacity challenge. We demonstrate for the first time adversarial testing-time attacks for this application for both classification and regression settings. We show that minor modifications to the camera image (an L-2 distance of 0.82 for one of the considered models) result in mis-classification of an image to any class of attacker's choice. Furthermore, our regression attack results in a significant increase in Mean Square Error (MSE) - by a factor of 69 in the worst case.",

keywords = "Adversarial machine learning, Deep neural networks, Self driving cars",

author = "Alesia Chernikova and Alina Oprea and Cristina Nita-Rotaru and Baekgyu Kim",

note = "Publisher Copyright: {\textcopyright} 2019 IEEE.; 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019 ; Conference date: 23-05-2019",

year = "2019",

month = may,

doi = "10.1109/SPW.2019.00033",

language = "English",

series = "Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "132--137",

booktitle = "Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019",

}

Chernikova, A, Oprea, A, Nita-Rotaru, C & Kim, B 2019, Are self-driving cars secure? Evasion attacks against deep neural networks for steering angle prediction. in Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019., 8844593, Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019, Institute of Electrical and Electronics Engineers Inc., pp. 132-137, 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019, San Francisco, United States, 23/05/19. https://doi.org/10.1109/SPW.2019.00033

Are self-driving cars secure? Evasion attacks against deep neural networks for steering angle prediction. / Chernikova, Alesia; Oprea, Alina; Nita-Rotaru, Cristina et al.
Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019. Institute of Electrical and Electronics Engineers Inc., 2019. p. 132-137 8844593 (Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Are self-driving cars secure? Evasion attacks against deep neural networks for steering angle prediction

AU - Chernikova, Alesia

AU - Oprea, Alina

AU - Nita-Rotaru, Cristina

AU - Kim, Baekgyu

PY - 2019/5

Y1 - 2019/5

N2 - Deep Neural Networks (DNNs) have tremendous potential in advancing the vision for self-driving cars. However, the security of DNN models in this context leads to major safety implications and needs to be better understood. We consider the case study of steering angle prediction from camera images, using the dataset from the 2014 Udacity challenge. We demonstrate for the first time adversarial testing-time attacks for this application for both classification and regression settings. We show that minor modifications to the camera image (an L-2 distance of 0.82 for one of the considered models) result in mis-classification of an image to any class of attacker's choice. Furthermore, our regression attack results in a significant increase in Mean Square Error (MSE) - by a factor of 69 in the worst case.

AB - Deep Neural Networks (DNNs) have tremendous potential in advancing the vision for self-driving cars. However, the security of DNN models in this context leads to major safety implications and needs to be better understood. We consider the case study of steering angle prediction from camera images, using the dataset from the 2014 Udacity challenge. We demonstrate for the first time adversarial testing-time attacks for this application for both classification and regression settings. We show that minor modifications to the camera image (an L-2 distance of 0.82 for one of the considered models) result in mis-classification of an image to any class of attacker's choice. Furthermore, our regression attack results in a significant increase in Mean Square Error (MSE) - by a factor of 69 in the worst case.

KW - Adversarial machine learning

KW - Deep neural networks

KW - Self driving cars

UR - https://www.scopus.com/pages/publications/85073167207

U2 - 10.1109/SPW.2019.00033

DO - 10.1109/SPW.2019.00033

M3 - Conference contribution

AN - SCOPUS:85073167207

T3 - Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019

SP - 132

EP - 137

BT - Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019

Y2 - 23 May 2019

ER -

Chernikova A, Oprea A, Nita-Rotaru C, Kim B. Are self-driving cars secure? Evasion attacks against deep neural networks for steering angle prediction. In Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019. Institute of Electrical and Electronics Engineers Inc. 2019. p. 132-137. 8844593. (Proceedings - 2019 IEEE Symposium on Security and Privacy Workshops, SPW 2019). doi: 10.1109/SPW.2019.00033

Are self-driving cars secure? Evasion attacks against deep neural networks for steering angle prediction

Abstract

Publication series

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this