Adversarial AI Attacks and Defenses in Intrusion Detection System for IoT

About this project

Project description

To facilitate applications such as smart home and smart manufacturing, energy-constrained devices are inter-connected through bandwidth-constrained communication protocols to form the internet-of-things (IoT). Due to such constraints, the IoT networks fail to employ conventional security protocols. Hence, on one hand, the IoT devices are vulnerable to illegal device access and inference of sensitive information; on the other hand, their users are prone to spoofing attacks through which an attacker can feed malicious data to the user. One resource-efficient way to secure an IoT network is to deploy comprehensively trained intrusion detection systems (IDS). However, the existing IDS have been shown to be vulnerable against modern artificial intelligence (AI)-assisted attacks that exploit specific vulnerabilities of the IDS.

In this project, we will address this challenge by developing an adversarial AI-driven framework to design a novel IDS which can detect AI-assisted cyber-attacks on IoT networks. This study will require extensive efforts along three research thrusts:

  1. We will first develop a systematic fuzzing-based method to discover the attack surfaces which can be exploited through modern AI techniques for evading the state-of-the-art IDS for IoT networks.
  2. We will then develop a novel framework that will utilize the concepts of adversarial ML to learn robust detection mechanisms for such attacks.
  3. Finally, we will utilize the developed framework to design an IDS which will not only detect the AI-assisted attacks with high detection rate, but also employ novel techniques to minimize the number of potential false alarms (false positive and negative). We will analyze the effectiveness of the proposed IDS by conducting experiments on simulated IoT networks as well as on a testbed with real-world IoT devices.

Outcomes

The project will deliver:

  1.  a systematic method to discover exploitable attack surfaces in IoT networks,
  2. a framework for designing adversarial AI-driven security mechanisms, and
  3. an intrusion detection system that can detect AI-assisted attacks on IoT networks.

Information for applicants

Essential capabilities

Strong background in computer science and engineering

Desireable capabilities

Knowledge in cyber security and artificial intelligence

Expected qualifications (Course/Degrees etc.)

Bachelors with honors or Masters

Project supervisors

Principal supervisors

UQ Supervisor

Associate professor Dan Kim

School of Information Technology and Electrical Engineering
IITD Supervisor

Assistant professor Vireshwar Kumar

Department of Computer Science and Engineering