Offene Abschlussarbeiten

This thesis aims to investigate the reliability and trustworthiness of traffic sign classifiers when subjected to occlusions. Utilizing the German Traffic Sign Recognition Benchmark (GTSRB) dataset, this research will focus on annotating the dataset with various levels and types of occlusions to evaluate if the predictions are still trustworthy. The primary objective is to assess the performance degradation of the classifier under different occlusion scenarios and to develop strategies to enhance its robustness. This study is crucial for improving the safety and reliability of autonomous driving systems where traffic signs might be partially obscured.

This thesis will investigate methods for detecting natural adversarial examples against ImageNet classifiers using classic computer vision techniques. Adversarial examples are inputs to machine learning models that are designed to cause the model to make a mistake. This project will utilize the Harder ImageNet Test Set (https://arxiv.org/abs/1907.07174) as an dataset for Natural Adversarial Examples. The primary objective is to explore and compare the effectiveness of traditional computer vision methods, such as histograms and SIFT (Scale-Invariant Feature Transform), in identifying these adversarial examples. The outcome of this research will enhance our understanding of model vulnerabilities and contribute to developing more robust machine learning systems.

This thesis will explore the effectiveness of different optimization strategies in the generation of adversarial patches. Adversarial patches are small, intentionally designed perturbations that can cause machine learning models, particularly in computer vision, to misclassify inputs. The primary objective of this research is to compare various optimization techniques, such as gradient-based methods, evolutionary algorithms, and reinforcement learning, to determine which methods are most effective and efficient in creating these patches. The outcome of this research could significantly enhance our understanding of model vulnerabilities and contribute to the development of more robust machine learning systems.

The alps see a surge of mountain biking as a recreational activity. This leads to frequent encounters of hikers and bikers on shared trails, but also to crashes between bikers due to, bad visibility in curves. In our previous work, we have investigated various scenarios and solutions, for example, a biker-to-hiker warning system, or a collision warning system for bike parks. Essential elements for these systems include localization of bikers in alpine environments, communication with near-range radio technologies like WiFi or BLE, but also suitable design of user interfaces and many more. Based on such earlier works (documented in theses and publications), we already identified various open challenges and possible future work that you can contribute to through a thesis or project. Please contact us to identify and define a suitable topic definition fitting your interests and previous experience. The overall project is collaboration between Ulm University and University of Trento.

This thesis aims to investigate the reliability and trustworthiness of traffic sign classifiers when subjected to occlusions. Utilizing the German Traffic Sign Recognition Benchmark (GTSRB) dataset, this research will focus on annotating the dataset with various levels and types of occlusions to evaluate if the predictions are still trustworthy. The primary objective is to assess the performance degradation of the classifier under different occlusion scenarios and to develop strategies to enhance its robustness. This study is crucial for improving the safety and reliability of autonomous driving systems where traffic signs might be partially obscured.

This thesis will investigate methods for detecting natural adversarial examples against ImageNet classifiers using classic computer vision techniques. Adversarial examples are inputs to machine learning models that are designed to cause the model to make a mistake. This project will utilize the Harder ImageNet Test Set (https://arxiv.org/abs/1907.07174) as an dataset for Natural Adversarial Examples. The primary objective is to explore and compare the effectiveness of traditional computer vision methods, such as histograms and SIFT (Scale-Invariant Feature Transform), in identifying these adversarial examples. The outcome of this research will enhance our understanding of model vulnerabilities and contribute to developing more robust machine learning systems.

An autonomous vehicle is equipped with a variety of sensors that produce large quantites of data which the vehicle uses to run a lot of different safety-critical functions, such as Cooperative Adaptive Cruise Control or Park Assist. In this thesis, we focus on the trust between the vehicle computer and other in-vehicle components that it relies upon to provide non-compromised data as input to different safety-critical functions. The goal of the thesis is to build a tool that will automate building of in-vehicular trust models based on a system model of a vehicle. A system model of a simplified vehicle will first need to be created by using the System Modeling Language (SysML). This model will serve as an input to the automation tool that needs to output a trust model in a pre-defined form. The methodology for building such trust models will be provided.

This thesis will explore the effectiveness of different optimization strategies in the generation of adversarial patches. Adversarial patches are small, intentionally designed perturbations that can cause machine learning models, particularly in computer vision, to misclassify inputs. The primary objective of this research is to compare various optimization techniques, such as gradient-based methods, evolutionary algorithms, and reinforcement learning, to determine which methods are most effective and efficient in creating these patches. The outcome of this research could significantly enhance our understanding of model vulnerabilities and contribute to the development of more robust machine learning systems.