REUSE me for a new project :-)

Potential supervisors: 

RISE - The Research Institute of Sweden has announced four possible thesis works within Security and Resilient Vehicle Design


Details can be found in the attached PDF, here follows a brief overiew of the topics:


Thesis 1: Hybrid and hierarchical intrusion detection in connected vehicles

An Intrusion Detection System (IDS) is an important weapon against cyber-threats in future vehicles. Unfortunately, vehicle electronics are not always capable of performing intrusion detection at an acceptable level. This is especially true for demanding IDS implementations (such as ones utilizing deep learning) or detections tasks that require data from multiple sources. 

An obvious solution would be to move some calculations to a cloud environment. However, this needs to be efficient to be practical. For example, the amount of data that must be sent between the vehicle and the cloud may become a bottleneck.


Thesis 2: Quantitative comparison of different approaches to intrusion detection in vehicles

Intrusion Detection Systems (IDS) are used to detect security anomalies in computers or networks of computers. Such systems can be implemented in different ways, each with different properties. 

For example, signature-based intrusion detection systems monitor certain system events and compare them against a database of known malicious patterns, which makes them highly deterministic and usually with only a small footprint (e.g. low memory and CPU usage). While IDS based on artificial intelligence (and deep learning in particular) have often higher complexity and harder to analyse but are sometimes able to detect previously unknown attacks.


Thesis 3: Resilient machine learning for vehicle intrusion detection

Intrusion Detection Systems (IDS) used in next generation connected vehicles sometimes use machine learning (and in particular deep learning) to detect cybersecurity threats. The models used in these systems often use a traditional purely mathematical approach to anomaly detection which assumes a static and non-malicious counterpart. However, this might not be the case for intrusion detection systems which engage a malicious attacker that is able to adopt.

In this thesis we want to investigate how vulnerable ML-based IDS implementations are to various adversarial attacks such as evasion attacks (i.e. mislead the prediction function) and poisoning attacks (i.e. manipulation of the learning process with crafted learning data). Furthermore, we want to examine how such attacks can be avoided or reduced by taking a security-aware approach to machine learning.


Thesis 4: Models for fleet-based intrusion detection

Automotive intrusion detection systems will in future to a certain extent be placed in cloud environments (e.g. fleet management systems). These systems use a model of the target vehicle together with data from vehicles to detect anomalies and security issues. To increase efficiency, simplify development and reduce maintenance costs, it would be good if such systems and corresponding models could be generalized to a few base vehicle models.


For more information and contact persons, please see the attached PDF.

Date range: 
December, 2020 to May, 2021