05 Fakultät Informatik, Elektrotechnik und Informationstechnik
Permanent URI for this collectionhttps://elib.uni-stuttgart.de/handle/11682/6
Browse
Search Results
Item Open Access Plug-and-play domain adaptation for neural machine translation(2023) Kadiķis, EmīlsNeural machine translation has emerged as a powerful tool, yet its performance heavily relies on training data. In a fast-changing world, dealing with out-of-domain data remains a challenge, prompting the need for adaptable translation systems. While fine-tuning is a proven effective adaptation method, it is not always feasible due to data availability, memory, and computational constraints. This thesis introduces a dynamic plug-and-play method inspired by controllable text generation to enhance machine translation across various domains without fine-tuning. This method, called Plug-and-Play Neural Machine Translation (PPNMT), uses a mono-lingual domain-specific bag-of-words to push the hidden state of the decoder through backrpopogation, making the output more in-domain. The method is tested on two types of domains: formality, gender (where the source language does not make a distinction between these aspects, but the target language does), and fine-grained technical domains (which are more based on topic inherent in the text on both the source and target sides). The method performs reasonably well for adapting the translation to different formality levels and, to a lesser extent, grammatical genders, even with an incredibly simple bag-of-words. However, it struggles with adapting the model to technical domains, and a fine-tuning baseline outperforms the proposed method in anything but very low few-shot settings in all tried domains. Despite that, the method shows some interesting behaviour, adapting to the formality on a level that goes beyond just using formal pronouns.Item Open Access Entwurf und Implementierung eines Stream-basierten, dynamischen und fairen Scheduling-Verfahrens für WiFi-Netzwerkverkehr(2023) Herrmann, JonaHeutzutage sind Wireless Local Area Network (WLAN)-Netzwerke, basierend auf dem IEEE~802.11 Standard, für mobile Endgeräte besonders wichtig, da diese nur so Zugang zum Internet bekommen können. Dementsprechend gibt es solche Netzwerke fast überall wie beispielsweise in öffentlichen Verkehrsmitteln, in Hotels oder auch bei der Arbeit. Doch die Qualität und Geschwindigkeit einer Verbindung kann dabei stark variieren. Um dies zu verbessern, wird zuerst eine Analyse dieser Netzwerke vorgenommen, um die Stelle des Bottlenecks zu identifizieren. Dabei konnte gezeigt werden, dass der Bottleneck an zwei Stellen vorliegen könnte. Das wäre einmal der Uplink zum Internet und andererseits der entsprechende Downlink, wobei sich der Bottleneck am Downlink letztendlich beim Internet-Provider befindet. Mit einem neu entwickelten Stream-basierten, dynamischen und fairen Scheduling-Verfahren soll die Qualität und Geschwindigkeit im WLAN-Netzwerk verbessert werden. Dafür wird eine neue Art von Fairness definiert, sodass die Pakete von Endgeräten mit einem geringen Datenverbrauch eine höhere Priorität erhalten. Dadurch bekommen letztendlich Endgeräte, welche gutmütig sind, eine bessere Antwortzeit als diese, die eine große Datenmenge übertragen. Zum Schluss wird die Linux-Implementierung des Scheduling-Verfahrens noch unter verschiedenen Metriken evaluiert. Dabei konnte gezeigt werden, dass die gewünschte Art von Fairness damit realisiert werden kann. Dies wurde sowohl unter Laborbedingungen als auch mit realen Anwendungen des Internets erfolgreich evaluiert. Außerdem wurde durch Messungen gezeigt, dass es im Punkt Performanz keinen signifikanten Unterschied zu dem Standard Scheduling-Verfahren in Linux gibt.Item Open Access Filter dictionaries for optical flow prediction with RAFT(2023) Walter, PeterIn the field of optical flow estimation, a dense vector field must be generated describing the apparent two-dimensional displacement of objects in consecutive images of a sequence. Although state of the art predictions are currently produced by deep convolutional neural networks, one major issue is that they are strongly susceptible to adversarial attacks, such as the Perturbation Constrained Flow Attack, which create small, noisy perturbations pursuing maximal change in the optical flow estimate. To improve adversarial robustness, this thesis includes receptive field convolutional layers into the optical flow predicting neural network RAFT. These receptive field layers use filter dictionaries to impose specific (geometric) priors onto convolutional kernels and improve results in image classification and reconstruction tasks. Each kernel in these RFCNNs can be written as a weighted sum over a fixed subset of filters taken from the dictionary. Besides the existing Gaussian derivative and Parseval completed sparse directional dictionaries, a novel PCA dictionary is proposed which consists of the principal components of the previously trained network’s kernels. All types of dictionaries are compared against each other at multiple positions in the network. Results show that receptive fields in individual layers mostly do not affect and in RAFT’s feature encoder even degrade performance, while Parseval completed dictionaries do not benefit the neural network in this context of optical flow. However, filter dictionaries with geometric motivations in RAFT’s update block, namely the Gaussian derivatives and sparse directional FDs, make the network up to 20% more robust against the PCFA in exchange for a worse fit in quality.Item Open Access OpenID for Verifiable Credentials : formal security analysis using the Web Infrastructure Model(2023) Hauck, FabianIn our increasingly connected world, digital identities play a fundamental role in delivering secure online services around the globe. To enable the seamless exchange of identification data among various entities, the adoption of standardized protocols is essential. The protocol family OpenID for Verifiable Credentials (OID4VC) is ideally suited for exchanging identities. The two most important protocols in this family are OpenID for Verifiable Credential Issuance (OID4VCI) and OpenID for Verifiable Presentations (OID4VP) with a wide range of applications in e-government as well as in the private sector. A prominent example is the European Digital Identity Framework, which includes these two protocols, among others. This means that any future wallet in the European Union will implement OID4VCI and OID4VP. Therefore, it is extremely important to guarantee their security. This thesis performs a rigorous formal security analysis of both the OpenID for Verifiable Credential Issuance and the OpenID for Verifiable Presentations protocols. In particular, we focus on analyzing the security of both protocols when they interact in an ecosystem. It is not sufficient to consider the two protocols separately, because the interaction between them may introduce new vulnerabilities. Therefore, the formal model created in this thesis models both protocols simultaneously. The model is based on the Web Infrastructure Model (WIM), which closely follows existing web technologies. To describe what security means in this context, we define an authentication security property and a session integrity security property for OID4VCI and OID4VP. We prove that the model is secure with respect to the security properties under the assumption of a vigilant user. If this assumption is violated, we have discovered a number of attacks. This work makes several contributions to the protocol specifications: First, the discovered vulnera- bilities were brought to the attention of the working group. Second, several issues were filed to improve the quality and security of the specifications. Lastly, we engaged in ongoing discussions on related issues.Item Open Access Exploration support for performance maps(2023) Galuschka, MarcelIn der architektonischen Entwurfsoptimierung ist die Analyse des mehrdimensionalen Parameterraums ein wichtiger Vorgang, um neue Lösungen zu finden. In dieser Arbeit wird ein glyphenbasierter Ansatz entwickelt, um die Informationen des Parameterraums und des kontinuierlichen Lösungsraums zu kombinieren. Die auf einem einheitlichen Gitter dargestellten Glyphen kodieren zusätzlich die Unsicherheit des zugrunde liegenden kontinuierlichen Raumes. Interaktion wird hinzugefügt, um Informationen mit höherer Auflösung zu erhalten. Eine Feedback-Schleife wird eingeführt, um Bereiche anzuzeigen, die bereits erforscht wurden. Die Anwendung des Tools wird demonstriert und in einer Expertenstudie bewertet. Die Ergebnisse der Studie werden klassifiziert und für technische Verbesserungen und zukünftige Forschung aufbereitet.Item Open Access GETACAR : a privacy-preserving platform for ride-pooling(2023) Hüppelshäuser, HansThe widespread adoption of autonomous vehicles is expected to lead to an overall increase in traffic. Ride-pooling can counter this downside of an otherwise promising technology, but the majority of current ride-pooling platforms utilise centralised designs that allow companies to collect vast amounts of user data. To solve this problem, we propose the decentralised ride-pooling platform GETACAR that focuses on privacy-preservation. GETACAR utilises blockchain technology to allow for the transparent and immutable tracking of ride processes without exposing personal information to other participants or the platform itself. To realise the platform, we develop its design, define its interactions and create a prototypical implementation. GETACAR is comprised of several components, including a customer and ride provider frontend allowing humans and autonomous vehicles to interact with GETACAR.We introduce an off-chain matching service to find the best possible match between customers and ride providers via a Vickrey auction. GETACAR also connects with crypto exchanges that allow the platform to use cryptocurrencies for internal transactions while users can still handle payments via fiat currencies. An authentication service verifies all parties wishing to participate on the platform, ensuring accountability and impeding the use of multiple identities. To ensure safety across the platforms, a robust rating system is in place that allows all parties to rate each other. In addition, a number of privacy mechanisms are in place to minimise the exposure of personal information, including location cloaking, pseudonyms, and frequently changing wallets. A prototype validates the GETACAR platform design by showcasing the platform’s key features, including a fully realised user frontend, the matching service, and smart contracts running on the Ethereum blockchain. All these components are connected and working together, allowing for a customer to request a ride with multiple ride providers bidding on the ride. The implemented matching service determines the winner, and smart contracts manage the overall ride, including the rating process. Both the design of the platform and the prototype showcase the potential of blockchain technology to create next-generation ride-pooling platforms that ensure transparency while preserving privacy.Item Open Access Algorithms for calculating robust schedules for time sensitive networking (TSN)(2023) Nieß, AdriaanWith the 802.1Qbv standard, the time-sensitive networking (TSN) task group has specified a TDMA-based scheduling method based on Ethernet, that allows to ensure real-time guarantees for time-critical data streams. A major challenge here is the computation of the cyclic schedules required for the switch configurations in the network. There exist already a number of different approaches to calculate these schedules. However, most of them focus on supporting a high number of streams and optimize target functions such as the Makespan. This leads to schedules that are particularly fragile against errors in the system model. To close this gap, we have developed a scheduler based on simple temporal networks (STNs) that maximizes the intervals allocated for the transmission of data packets in order to ensure a very high degree of robustness against unforeseen delay or cross-traffic. In a simulation it could be shown that the calculated schedules could still guarantee a loss-free data transmission under compliance with all deadlines if the actual per-hop network delay deviated from the assumed delay by a factor of up to 7. Furthermore, an even larger error did not immediately lead to a breakdown, but to a gradual degradation. This makes the scheduling method presented here as well as the underlying algorithms particularly interesting for applications that come with a high degree of uncertainty in the system model, such as wireless communications or heterogeneous networks.Item Open Access Domain-driven monitoring of business-critical application systems(2023) Wagner, LionContext. Many software companies use application performance management (APM) to track the technical healthiness of their systems. Modern frameworks, that are used to instrument such systems, like OpenTelemetry or AppDynamics, allow for an easy observation of many technical metrics, such as response times, throughout or boot times. Problem. However, underneath each systems lies usually a complex business process with its own non-technical characteristics, so called domain metrics. These characteristics are often not directly observable by technical metrics, which makes them hard to measure and understand, resulting in a conceptual gap between developers and domain experts. Objective. This thesis presents the domain-driven instrumentation (DDI) approach, which eases the process of instrumenting a software system to measure domain metrics, by automatically injecting instrumentation points into a running system, based on a provided domain-architecture-mapping (DAM). Additionally, the approach is tested and demonstrate on a realistic mock system, to keep the results representative. Method. Based on these objectives, first the mock system, which is strongly based on real world industry system was developed. Then, based on the work of Hofer and Schwentner [HS21b] and Vernon [Ver13] a meta-model, that can be used to model a DAM, based on the concepts of domain-driven design (DDD) and domain storytelling (DST), was developed. This model was then implemented in to the dqualizer tool chain, and executed on the mock system, to show its feasibility. Result. The demonstration revealed, that this approach can be effectively used to instrument a modern software system for monitoring domain metrics, by successfully instrumenting the mock with tracing and metric monitoring probes. However, it also revealed some major open problems, when it comes to instrumenting asynchronous behavior on a domain level, resulting in the collection, but non-usability of some metrics. Conclusion. In conclusion, the work of the thesis was able to successfully implement and demonstrate an automated domain-driven instrumentation approach and demonstrate it on a system that is closely related to a real world setting.Item Open Access Investigation of multistability-affected period-incrementing and period-adding structures in a model of a power converter with symmetry(2023) Weik, FabianAlthough the established theory of non-linear dynamical systems is effective at describing physical systems and phenomena, the theory has been developed with a focus on smooth systems, limiting its applicability in certain areas. This limitation becomes apparent when modeling piecewise-smooth systems such as electrical systems that contain at least one switching element or mechanical systems with collisions. This thesis deals with a model of a DC/AC power converter that is piecewise-smooth, discontinuous, and has a certain symmetry. The definition of this model is exceptionally complex, and the model exhibits an unusual period-incrementing structure that is affected by multistability. This thesis identifies the characteristics of the model that lead to this unusual bifurcation structure by constructing an archetypal model that exhibits the same bifurcation behavior. It follows a description of the dynamics of the archetypal model and an explanation of the bifurcation structure using the description of the dynamics. Additionally, this thesis demonstrates that the proposed archetypal model can exhibit behavior leading to bifurcation structures that are related to period-adding structures. The resulting bifurcation structures behave unexpectedly, and this behavior is explained by leveraging the symmetry present in both the archetypal and the original model. Using the symmetry present, the mathematical rules governing the resulting bifurcation structures are derived from the rules governing classical period-adding structures.Item Open Access Gender bias in dependency parsing(2023) Go, Paul StanleyRecent high-profile advances in natural language processing (NLP) have spurred interest into identifying and rectifying socially harmful problems common in NLP systems such as gender bias. Unfortunately, many works which attempt to tackle the issue of gender bias suffer from methodological deficiencies such as the assumption of a binary and immutable concept of gender. We scrutinize one such work which found gender bias in dependency parsing and evaluate if the claims have merit. Our results were inconsistent with the gender bias findings of that paper, and further investigations through error analysis and treebank analysis revealed methodological flaws which artificially introduced differences between their female and male data sets. Mistakes made during preprocessing compromised the outcome; therefore, their results do not prove the existence of gender bias in dependency parsing. Through our findings, we suggest a different methodology for identifying and alleviating syntactic bias that is more inclusive for everyone-no matter their gender.