Die vorliegende Arbeit vergleicht die Eignung von Flutter und Kotlin Multiplatform für den Relaunch einer mobilen App. Angesichts der Vielzahl von Entwicklungs-Frameworks ist die Auswahl des richtigen Frameworks von entscheidender Bedeutung. Diese Thesis führt in die Welt der Cross-Plattform-Entwicklung ein, um zu ergründen, welches Framework am besten geeignet ist. Die Motivation für die Untersuchung liegt in der hohen Relevanz mobiler Plattformen („Mobile First”) und der Notwendigkeit, eine bestehende Anwendung neu zu entwickeln, um mit den aktuellen Entwicklungsstandards Schritt zu halten. Die wissenschaftliche Fragestellung dieser Arbeit konzentriert sich darauf, wie sich die Eignung eines Cross-Plattform-Frameworks für die Neuentwicklung einer bestehenden Anwendung strukturiert und allgemeingültig evaluieren lässt. Die Ergebnisse sollen eine Entscheidungsmatrix für den allgemeinen Einsatz sowie einen konkreten Vorschlag für den Fall der „mQuest”-App liefern. Die Arbeit erläutert zunächst die Grundlagen der Cross-Plattform-Entwicklung, betrachtet die relevanten Frameworks und Technologien sowie bisherige Arbeiten auf diesem Gebiet. Im weiteren Verlauf werden nur noch die in einer Vorauswahl ausgewählten Frameworks Flutter und Kotlin Multiplatform betrachtet. Das Resultat der Entscheidungsmatrix offenbart für Flutter einen leichten Vorsprung aufgrund seiner Reife und Verfügbarkeit gut dokumentierter Pakete. Kotlin Multiplatform muss hier noch aufholen. In der abschließenden Diskussion wird erörtert, wie ein Migrationspfad für den Relaunch der mQuest App aussehen könnte und welche Herausforderungen dabei zu beachten sind. Dazu zählen die Wartung der bestehenden Lösung und die Vermeidung von Regressionsfehlern in der neuen Software.

In the field of product matching, accurately matching product images and descriptions has proven to be a great challenge due to unreliability and inconsistencies on both the vendor and customer side. While unimodal methods, which only process one modality such as image or text, have been researched extensively, they often fail to provide a complete understanding of the important product features. To address this, multimodal deep learning has been a rising field in machine learning in recent times, which aims to combine information from multiple modalities, such as image, text and audio, in order to capture all critical product information. Inspired by recent advances in this field, this thesis employs a multimodal neural network based on a bidirectional triplet loss function, which maps similar image and text embeddings closer to each other in an embedding space. As a backbone, the model uses a Convolutional Neural Network on both the image and text side. What sets this thesis apart as particularly innovative is that a character-level approach is used on the text side as backbone, instead of common word embedding methods. This is further supported by recent achievements in multimodal deep learning, demonstrating that the character-level approach is particularly effective. The research in this thesis is conducted in collaboration with Parsionate GmbH, an IT consulting company specializing in data management, located in Stuttgart. Parsionate provides a HORNBACH dataset consisting of DIY products for experimentation in this thesis. During the experiments, the optimal hyperparameters are identified using a grid search approach. The experimental results highlight that the multimodal neural network with bidrectional triplet loss is able to outperform unimodal methods when evaluated directly against the top 5 similar products. Notably, the multimodal network with a character-level Convolutional Neural Network for text processing and ResNet50 for image processing outperforms all other word embedding methods. These findings strongly suggest that further investigating multimodal neural networks with character-level approaches opens up new avenues for research and subsequent application in the product matching domain.

Securing software is one of the most important parts in modern software development. Fuzzing has become one of the most popular methods to automatically test software. Most fuzzing approaches need the target software to be recompiled which presupposes source code to be available. When no source code is available, black box fuzzers are used. In modern software, states play a big role in its functioning. A black box fuzzer can come to its limits quick when operating on a stateful target with no knowledge. The use of a state machine in a fuzzer can make the fuzzer more effective. This thesis introduces a state machine estimation tool for black box systems. An approach to estimate the state machine with state-of-the-art algorithms over a defined interface is proposed. Fuzzing will be used to find more inputs and states of the target to make a more complete state machine. The implemented approach is evaluated on two stateful targets LightFTP and BFTPD. With a set of pre-known inputs, the tool was able to correctly estimate the state machines of the targets and the fuzzing method proved to be successful in finding more states and inputs. Multiple fuzzing techniques and automata learning algorithms were benchmarked to find the most successful combination.