Publications

BibTex format

@inproceedings{Zizzo:2020,
author = {Zizzo, G and Hankin, C and Maffeis, S and Jones, K},
publisher = {Institute of Electrical and Electronics Engineers},
title = {Adversarial attacks on time-series intrusion detection for industrial control systems},
url = {http://hdl.handle.net/10044/1/84057},
year = {2020}
}

Download

RIS format (EndNote, RefMan)

TY  - CPAPER
AB  - Neural networks are increasingly used for intrusiondetection on industrial control systems (ICS). With neuralnetworks being vulnerable to adversarial examples, attackerswho wish to cause damage to an ICS can attempt to hidetheir attacks from detection by using adversarial exampletechniques. In this work we address the domain specificchallenges of constructing such attacks against autoregressivebased intrusion detection systems (IDS) in a ICS setting.We model an attacker that can compromise a subset ofsensors in a ICS which has a LSTM based IDS. The attackermanipulates the data sent to the IDS, and seeks to hide thepresence of real cyber-physical attacks occurring in the ICS.We evaluate our adversarial attack methodology on theSecure Water Treatment system when examining solely continuous data, and on data containing a mixture of discrete andcontinuous variables. In the continuous data domain our attacksuccessfully hides the cyber-physical attacks requiring 2.87 outof 12 monitored sensors to be compromised on average. Withboth discrete and continuous data our attack required, onaverage, 3.74 out of 26 monitored sensors to be compromised.
AU  - Zizzo,G
AU  - Hankin,C
AU  - Maffeis,S
AU  - Jones,K
PB  - Institute of Electrical and Electronics Engineers
PY  - 2020///
TI  - Adversarial attacks on time-series intrusion detection for industrial control systems
UR  - http://hdl.handle.net/10044/1/84057
ER  -

Download

DrSergioMaffeis

Contact

Location

Summary

Citation

BibTex format

RIS format (EndNote, RefMan)