Contributions published at Communication Systems Group (Burkhard Stiller)

As recently reported by the World Health Organization (WHO), the high use of intelligent devices such as smartphones, multimedia systems, or billboards causes an increase in distraction and, consequently, fatal accidents while driving. The use of EEG-based Brain–Computer Interfaces (BCIs) has been proposed as a promising way to detect distractions. However, existing solutions are not well suited for driving scenarios. They do not consider complementary data sources, such as contextual data, nor guarantee realistic scenarios with real-time communications between components. This work proposes an automatic framework for detecting distractions using BCIs and a realistic driving simulator. The framework employs different supervised Machine Learning (ML)-based models on classifying the different types of distractions using Electroencephalography (EEG) and contextual driving data collected by car sensors, such as line crossings or objects detection. This framework has been evaluated using a driving scenario without distractions and a similar one where visual and cognitive distractions are generated for ten subjects. The proposed framework achieved 83.9% -score with a binary model and 73% with a multiclass model using EEG, improving 7% in binary classification and 8% in multi-class classification by incorporating contextual driving into the training dataset. Finally, the results were confirmed by a neurophysiological study, which revealed significantly higher voltage in selective attention and multitasking.

Switzerland allows its citizens to participate in elections and votes by casting ballots via postal mail. This practice, referred to as “Remote Postal Voting”, is a convenient way for voters to cast their ballots without visiting a polling station. However, the application of Remote Postal Voting has raised security concerns regarding vote- and election manipulations by voting officials or third parties. These concerns have been confirmed in practice, where malicious activities have been discovered on multiple occasions. This thesis aims at providing verifiability for Remote Postal Voting procedures by presenting the design and implementation of a verifiable Remote Postal Voting system. The design integrates well into the existing Swiss Remote Postal Voting procedures. It leverages blockchain technology, Homomorphic Encryption, Non-Interactive Zero Knowledge Proofs, and Threshold Cryptography to meet privacy and verifiability requirements. The evaluation shows that the proposed system scales well in the Swiss Remote Postal Voting scenario, whereas further optimization would be necessary for larger countries.

Cryptographic ransomware encrypts files and demands a ransom for their decryption. Ransomware is increasingly targeting Internet of Things (IoT) devices that contain critical data. Due to limited resources, IoT devices cannot implement resource-intensive protection mechanisms to defend against ransomware. To provide a lightweight ransomware protection mechanism for IoT devices, three overlay file systems have been implemented. The overlay file systems use moving-target defense techniques to hide file type identification, increase encryption time, and trap ransomware in infinite directories. The evaluation results show that the implemented overlay file systems provide protection against ransomware attacks. The main limitations of the overlay file systems are the inability to distinguish between malicious and non-malicious applications and the performance overhead for small file sizes.

The explosive growth of the IoT has come along with an increase of cyberattacks with ransomware, rootkits and Command-and-Control malware being particularly common families. One promising approach for mitigation is offered by Moving Target Defense (MTD), which works by dynamically altering a target’s attack surface. However, the state of IoT MTD is still immature, especially lacking research dedicated to coordinating multiple MTD techniques in real applications. As a means to optimize such a system, this work explores the application of reinforcement learning (RL) to reactively deploy MTD techniques against the aforementioned malware families in a real crowdsensing scenario. First, the task of RL-based MTD selection is analyzed to distill major system requirements. Thereafter, three training simulations are presented along with the implementation of a complete, online MTD agent. As online RL is costly, the simulations gradually shift from a rather theoretical perspective towards approximating reality to allow policy transfer to a real environment. Using a supervisor to create reward signals, the first simulation marks a baseline. The second exchanges this supervisor for an anomaly detection component. For comparability both simulations use a new dataset of raw attack behaviors. The third simulation also leverages anomaly detection, yet utilizes a second dataset of behaviors monitored by a real online agent. While the agent of the first simulation learns to select MTD techniques against all attacks of the aforementioned families, the second and third simulations show that a realistic agent’s convergence is affected by anomaly detection inaccuracies, but generally attacks are effectively mitigated. Finally, implications of the online agent are discussed and its resource consumption is evaluated on a Raspberry Pi 3. Requiring less than 1MB storage and always utilizing below 80% of the available CPU and RAM, hardware poses no limitation. However, the time required to learn new attacks may impair viability.

null

Internet-of-Things (IoT), Artificial Intelligence (AI), and Blockchains (BCs) are essential techniques that are heavily researched and investigated today. This work here specifies, implements, and evaluates an IoT architecture with integrated BC and AI functionality to manage access control based on facial detection and recognition by incorporating the most recent state-of-the-art techniques. The system developed uses IoT devices for video surveillance, AI for face recognition, and BCs for immutable permanent storage to provide excellent properties in terms of image quality, end-to-end delay, and energy efficiency.

null

null

null

Blockchains (BC) and Distributed Ledgers (DL) offer favorable properties, especially immutability and decentralization, which are suitable for voting systems’ Bulletin Boards (BB). In recent years, an influx of BC-based voting systems have been observed. Distributing trust among multiple trustees is a crucial reason to adopt BCs and DLs in voting systems. The practical deployment must be decentralized, too, and not just done through virtualizing interconnected systems. As discussed widely, adopting a BC or DL can incur threats to a system that assumed a trusted and centralized Public Bulletin Board (PBB). Therefore, the exploitation of BCs or DLs requires careful consideration of cryptographic mechanisms and the overall system design, as well as the adversary model. Besides these operational necessities, the long-term privacy of ballots is essential. Thus, the key question investigated in this article is: Can BC, and DL-based voting systems be considered harmful? Hence, first (i) requirements of BC-based voting systems are provided, followedby (ii) terminology definitions, and (iii) complemented by the design and implementation of a fully decentralized voting system: Æternum, which achieves Unconditional Privacy (UP) and neither relies on computational hardness assumptions nor on a trusted Trusted Third Party (TTP). Achieving UP is crucial because future adversaries may be able to break hardness assumptions. Æternum does not present a Single-Point-of-Failure (SPoF) either, since (i) the PBB in use is based on a permissioned DL, and (ii) the final tally and proofs can be verified by anyone, without requiring trust in any authority.

IoT technologies have experienced a continuous upswing in recent years, resulting in countless types of networked devices. With the flourishing of these technologies and especially the use of IoT sensors, daily life has been improved in many ways and brought new opportunities, such as the IoT crowdsensing platform ElectroSense. The fact that these resource-constraint devices are vulnerable to cyberattacks raises the question of how to provide a suitable and reliable security system. Using the MTD paradigm suggests deploying specific defense mechanisms to mitigate malicious software damage. For the actual decision process, i.e., under which circumstances a corresponding countermeasure is triggered, often complex technologies such as machine learning, game theory, or evolutionary algorithms are used based on non-system metrics. This work contributes to the research by proposing an MTD Strategy Selection Agent StraSelA that makes decisions using a simple set of rules, also called policy, based on system metrics. For this purpose, seven malware from Command and Control, Ransomware, and Rootkit, as well as an ElectroSense remote sensor running on a Raspberry Pie 4, were considered. The underlying metrics were initially defined in systematic literature research about malware. Subsequently, the device behavior was recorded, and specific rules for executing defense measures were created through data analysis of the individual metrics and then experimentally fine-tuned. The architecture of StraSelA was defined and then implemented so that the software, together with the synthesized policy database on the IoT device, can firstly detect any malware that occurs and secondly trigger a suitable countermeasure accordingly. After evaluating various metrics, such as detection rate and overhead, it was shown that it could protect a system from attackers with a simple and resource-efficient decision-making process based on system metrics. Thus, the added value coming from the simple strategy selection algorithm of StraSelA is confirmed. Finally, all data sets used during the policy synthesis and evaluation are freely available for further research.

Systems utilizing artificial intelligence (AI) are becoming more and more useful in supporting human decision-making tasks. Unsupervised AI algorithms play a significant role in the detection of intrusions and cyberattacks on devices with limited resources. However, current solutions focus on achieving the best detection performance, missing the importance of quantifying the trustworthiness level of the trained models and their predictions. This work focuses on computing the level of trustworthiness for unsupervised anomaly detection models. In this work a taxonomy with four different pillars of trust and associated metrics is proposed. Further, an algorithm has been developed which takes unsupervised anomaly detection models together with the underlying training, test and outlier data sets as inputs to compute an overall trust score. This algorithm has further on been embedded in a web application which is available to model developers and serves as a global solution to evaluate unsupervised anomaly detection models towards trustworthiness. Lastly, an in-depth analysis on different models has been conducted in order to evaluate the proposed algorithm and to point out strengths and limitations.

As the significance of internet services grows, the need to maintain their online presence likewise grows. During Covid-19, we witnessed the significance of digitization in our daily lives. For example, in Education, despite the crisis, students were able to attend online courses and take tests. But also in several other vital disciplines. However, relying only on digital gadgets to provide a service is problematic since we will be unable to complete the intended work if they go down. There are several reasons why an internet service might go down, including a power outage, faulty source code, or a cyber assault. GridDB focuses on a well-known issue in computer networks, Distributed Denial of Service (DDoS) attacks. DDoS attacks are a significant threat to Internet availability that has not been resolved despite several commercial and academic initiatives. DDoS attacks occur on a daily basis, and as society continues to digitize, these attacks offer several dangers to organizations and people. In previous work, the DDoS Clearing-house system was built on sharing DDoS fingerprints, a signature of DDoS attacks registered by partners in a cooperative alliance to counter DDoS attacks. GridDB provides a straightforward Frontend to manage handling packet captures (PCAPs) and sharing information. Thus, the Frontend should interface with DDoS Clearing-House to update its status. Therefore, portable Backend and Frontend applications were implemented. The primary outcome of this work is the integration of the DDoSCH components into one fully automated and portable system, replacing the terminal-based application provided by the DDoS Clearing-House with a user-friendly UI. The implemented application allows users to view analyzed attack data in various visualizations to increase the perception and understanding of the attack nature, which gives users the ability to mitigate similar attacks in the future and to protect their potential infrastructure.

In today’s world, open-source software components are present in almost any codebase. However, the majority of these codebases also contain outdated components, which can lead to problems. To mitigate this, there has been a shift toward financially funding opensource projects to increase developers’ commitment to maintaining the projects. However, the majority of sponsorship platforms focus on bringing funds to the projects and leave the decision of how to distribute it among the developers to the project owners. This thesis provides a method for automatically estimating a developer’s contribution to an open-source GitHub repository. A contribution engine is developed that takes lines of code as a baseline for the contribution and considers the metrics maintainability, bug fixes, and test coverage to add a bonus and allows for customizability by adapting the importance of the metrics. The engine is implemented with the help of a university project and tested against a similar project to demonstrate its legitimacy. It was concluded that the proposed evaluation engine succeeds in fairly evaluating code contribution in the considered projects, but the metrics are highly subjective and thus cannot be generalized to different projects. Furthermore it was found that entirely automating the evaluation process was not possible within the considered definition of fairness.

The recent COVID pandemic posed a challenge to healthcare systems worldwide, exposing limited resources and inefficiency in managing PMDs (Portable Medical Devices). This thesis reports the prototypical implementation of an new and cost-efficient type of asset tracking architecture, enabling hospitals to efficiently locate their resources and maximize the usage of PMDs in emergency situations, such as pandemics or natural disasters, when a sudden increase in demand may be anticipated. PMD-Track’s key novelty lies in the use of staff’s smartphones to replace expensive stationary gateways, typically required in traditional inventory tracking systems. As employees approach tagged PMDs in their day-to-day, their smartphone updates the location of spotted PMDs in a real-time inventory system, providing room-level localization data with up to 95% accuracy. Experiments were evaluated by contrasting two techniques (fingerprinting and multi-lateration) based on two different test locations in Zurich.

null

Tactical Battlefield Networks are networks used by the military to establish communication between units on the field. These networks pose diverse environmental challenges that need to be addressed. They are organized in so-called Mobile Ad Hoc Networks, that mimic a civilian cellphone network infrastructure. Testing performance in these networks proves more difficult than testing traditional fixed communication networks. This thesis focuses on the development and implementation of an automated testing environment. This testing environment provides the user with the ability to convert arbitrary network topologies to a virtual environment. Further, traffic can be configured to automatically run on these networks and results can easily be fetched without the user directly having to interact with the virtual environment. To assess the state of the implementation, first experiments with a small network topology were conducted and the results of these experiments was shown and discussed.

It is of central importance for a democracy that all citizens entitled to vote have the opportunity to cast their ballots in votes and elections. The most common way to vote in Switzerland is to hand in a completed ballot directly at the urn or by postal services. However, an important part of Switzerland's e-government strategy is to enable voting and elections electronically. In this context, the Federal Council emphasizes that e-voting should not only be made available to the Swiss abroad, but to all citizens who are entitled to vote. For this reason, a new legal basis has come into force on 25.05.2022, which allows the cantons to offer e-voting as part of a trial operation. Numerous challenges must be overcome in the development of the digitalization of voting and elections. While there is already a large amount of scientific literature on this topic, most of it takes a centralized approach. The Provotum project, on the other hand, takes a different approach. Due to the use of distributed ledger technology, this electronic voting and election system is decentralized. This has desirable consequences as it allows the system to ensure transparency, integrity and robustness. In the latest version of Provotum, established cryptographic techniques have been included in the project, resulting in the enabling of elections and the shifting of computational effort to the election infrastructure. However, this version does not yet use a graphical user interface. The thesis implements graphical user interfaces for vote casting and for vote administration. This means that relevant literature on user interfaces and user experiences in other electronic voting systems will also be examined, and the Provotum 3.0 Mixnet prototype will also be analyzed. Based on this analysis, three user interfaces will be created so that each stakeholder in Provotum can complete their tasks over the Internet. In addition, it creates a simple API that enables communication between the blockchain and the graphical user interfaces. These software components are implemented using modern technologies. Finally, user interfaces are evaluated using heuristics, use cases, a discussion and system usability scales.