Titulo Estágio
Adversarial Techniques for the Evaluation and Improvement of Intrusion Detection Systems
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
CISUC-SSE
Enquadramento
In recent years there has been a proliferation of internet-based services. As a result, the number of users and network traffic has increased considerably. Given the criticality and high value of services provided, it attracts a lot of malicious attacks. A successful attack can have serious consequences and violate existing security policies (i.e., Confidentiality, Integrity, Availability (CIA) [1]). Various types of attacks exist, from password cracking [2] (e.g., John the Ripper [3]) and IoT malware (e.g., Mirai Botnet [4]) which compromise confidentiality; man-in-the-middle attacks (e.g., ARP spoofing [5]) which violate both confidentiality and integrity; Denial of Service (DoS) attacks (e.g., HTTP flooding [6]) that compromise availability.
Intrusion Detection Systems (IDS) monitor the system or network looking for signs of intrusion. Host-based IDSs focus on a single system while Network IDSs monitors the whole system. A typical IDS approach consists of three phases: monitoring, detection, and response [7][14][15].
Given the recent success and applicability of Machine Learning (ML) algorithms, nowadays most IDS solutions also make use of them [2][7]. Using data collected from the system being monitored, ML algorithms try to identify whether an intrusion is ongoing. IDS can be further divided into signature-based (e.g., Snort[8] and Suricata [9]) and anomaly-based (e.g., Zeek/Bro [10]).
ML solutions have achieved remarkable results in a variety of problems and domains. Notwithstanding, it has also been shown that they have significant limitations that open them to exploitation. State-of-the-art algorithms, from tree-based to Deep Learning have been shown to be susceptible to adversarial attacks (e.g., [11]). Adversarial attacks are samples that have been crafted to fool ML models (e.g., an attack “disguised” to be classified as benign). Different types of adversarial attacks exist [12]: error-generic (i.e., misleading classification regardless of the output class) and error- specific (i.e., misclassifying as a specific output class) evasion attacks. Different levels of attacker’s knowledge can also be considered, depending on how much it knows about the system (i.e., white, gray, and black-box). Over the years various solutions have been proposed to address this, also known as defenses. In recent years some research has also been done focused on IDSs [13].
This internship consists of exploring the state-of-the-art of using adversarial techniques in the context of IDSs. The goal is to study existing work on both IDS and adversarial ML from a security-sensitive perspective and assess how they can be used for the evaluation and improvement of IDSs. To achieve this, this work will explore the use of existing intrusion datasets [16].
[1] Avizienis, A., et al. (2004). Basic concepts and taxonomy of dependable and secure computing. IEEE TDSC, 1(1), 11-33.
[2] He, K., Kim, D. D., & Asghar, M. R. (2023). Adversarial Machine Learning for Network Intrusion Detection Systems: A Comprehensive Survey. IEEE Communications Surveys & Tutorials.
[3] “Johntheripperpasswordcracker.”Openwall.2019. [Available: https://www.openwall.com/john
[4] Anna-Senpai, “Mirai source code.” 2017. Available: https://github.com/jgamblin/Mirai-Source-Code
[5] Y. Said. “ARP spoofing using a man-in-the-middle attack.” 2020 Available: https://linuxhint.com/ arp_spoofing_using_man_in_the_middle_attack
[6] “LOIC.” NewEraCracker. 2019. Available: https://github.com/NewEraCracker/LOIC
[7] Khraisat, A., et al (2019). Survey of intrusion detection systems: techniques, datasets and challenges. Cybersecurity, 2(1), 1-22.
[8] “Snort.. Available: https://www. snort.org
[9] “Suricata.” The Open Information Security Foundation. 2020. Accessed: Feb. 11, 2022. [Online]. Available: https://suricata-ids.org
[10] “Zeek.” 2020. Accessed: Feb. 11, 2022. Available: https:// zeek.org
[11] Papernot, N., Mcdaniel, P., and Goodfellow, I. (2016). Transferability in ma- chine learning: from phenomena to black-box attacks using adversarial samples. ArXiv, abs/1605.07277.
[12] Biggio, B. and Roli, F. (2018). Wild patterns: Ten years after the rise of ad- versarial machine learning. Pattern Recognition, 84:317–331.
[13] Alotaibi, A., & Rassam, M. A. (2023). Adversarial machine learning attacks against intrusion detection systems: A survey on strategies and defense. Future Internet, 15(2), 62.
[14] Yi, L., Yin, M., & Darbandi, M. (2023). A deep and systematic review of the intrusion detection systems in the fog environment. Transactions on Emerging Telecommunications Technologies, 34(1), e4632.
[15] Chang, V., Golightly, L., Modesti, P., Xu, Q. A., Doan, L. M. T., Hall, K., ... & Kobusińska, A. (2022). A survey on intrusion detection systems for fog and cloud computing. Future Internet, 14(3), 89.
[16] Khraisat, A., Gondal, I., Vamplew, P., & Kamruzzaman, J. (2019). Survey of intrusiondetection systems: techniques, datasets and challenges
Objetivo
The learning objectives of this master internship are:
1) Dependability, Security: study the subject of Dependability and Security and its attributes (confidentiality, integrity, and availability)
2) Intrusion Detection System (IDS): understand the problem of IDS, its various approaches, advantages and disadvantages, and practical applications
3) Machine Learning: understand how to use and the impact of ML techniques to support IDS
4) Adversarial Machine Learning: study the advanced topic of Adversarial ML focused on security-based domain
5) Research Design: understand how to design and execute an experimental process to address complex and open research issues
Plano de Trabalhos - Semestre 1
[11/09/2023 a 15/10/2023] Literature review
Study the concepts to be used in the internship, namely online dependability, security, machine learning, intrusion detection systems, and adversarial machine learning
[16/10/2023 a 05/11/2023] Analysis and selection of target techniques
Identification, analysis, and selection of which intrusion detection and machine learning techniques will be used, IDS datasets and intrusion injection approaches
[06/11/2023 a 03/12/2023] Definition of the experimental process
Design and plan the experimental process that will be used to conduct the study. This includes defining all the relevant components, from the approaches used for the intrusion detection systems, the machine learning techniques that will be used to support the detection, the workload/datasets that will be used to exercise the detection systems, the adversarial approaches to be explored, as well as the architecture of the testbed that will be used
[04/12/2023 a 15/01/2024] Write the dissertation plan
Plano de Trabalhos - Semestre 2
[05/02/2024 a 31/03/2024] Set up the experimental testbed
Set up the testbed required to conduct the experiments.
[01/04/2024 a 21/04/2024] Conduct the experimental campaign
Use the testbed to conduct the experimental process,
[22/04/2024 a 05/05/2024] Explore and assess the generated data
Process, explore and analyze the the results obtained from the experimental process on the use/impact of adversarial ML on the state-of-the-art of IDS
[06/05/2024 a 03/06/2024] Write the thesis.
Condições
Depending on the evolution of the internship a studentship may be available to support the development of the work. The work is to be executed at the laboratories of the CISUC’s Software and Systems Engineering (SSE) Group.
Orientador
João Rodrigues de Campos
jrcampos@dei.uc.pt 📩