Refine
Year of publication
- 2023 (25) (remove)
Document Type
- Article (8)
- Conference Proceeding (5)
- Master's Thesis (4)
- Bachelor Thesis (3)
- Course Material (2)
- Preprint (2)
- Working Paper (1)
Is part of the Bibliography
- no (25)
Keywords
- CI/CD (2)
- Chatbot (2)
- DevOps (2)
- Führung (2)
- Künstliche Intelligenz (2)
- Machine Learning (2)
- Maschinelles Lernen (2)
- Nachhaltigkeit (2)
- Psychology at Work (2)
- Selbstführung (2)
Institute
- Fakultät IV - Wirtschaft und Informatik (25) (remove)
In the last years generative models have gained large public attention due to their high level of quality in generated images. In short, generative models learn a distribution from a finite number of samples and are able then to generate infinite other samples. This can be applied to image data. In the past generative models have not been able to generate realistic images, but nowadays the results are almost indistinguishable from real images.
This work provides a comparative study of three generative models: Variational Autoencoder (VAE), Generative Adversarial Network (GAN) and Diffusion Models (DM). The goal is not to provide a definitive ranking indicating which one of them is the best, but to qualitatively and where possible quantitively decide which model is good with respect to a given criterion. Such criteria include realism, generalization and diversity, sampling, training difficulty, parameter efficiency, interpolating and inpainting capabilities, semantic editing as well as implementation difficulty. After a brief introduction of how each model works on the inside, they are compared against each other. The provided images help to see the differences among the models with respect to each criterion.
To give a short outlook on the results of the comparison of the three models, DMs generate most realistic images. They seem to generalize best and have a high variation among the generated images. However, they are based on an iterative process, which makes them the slowest of the three models in terms of sample generation time. On the other hand, GANs and VAEs generate their samples using one single forward-pass. The images generated by GANs are comparable to the DM and the images from VAEs are blurry, which makes them less desirable in comparison to GANs or DMs. However, both the VAE and the GAN, stand out from the DMs with respect to the interpolations and semantic editing, as they have a latent space, which makes space-walks possible and the changes are not as chaotic as in the case of DMs. Furthermore, concept-vectors can be found, which transform a given image along a given feature while leaving other features and structures mostly unchanged, which is difficult to archive with DMs.
The digital transformation with its new technologies and customer expectation has a significant effect on the customer channels in the insurance industry. The objective of this study is the identification of enabling and hindering factors for the adoption of online claim notification services that are an important part of the customer experience in insurance. For this purpose, we conducted a quantitative cross-sectional survey based on the exemplary scenario of car insurance in Germany and analyzed the data via structural equation modeling (SEM). The findings show that, besides classical technology acceptance factors such as perceived usefulness and ease of use, digital mindset and status quo behavior play a role: acceptance of digital innovations, lacking endurance as well as lacking frustration tolerance with the status quo lead to a higher intention for use. Moreover, the results are strongly moderated by the severity of the damage event—an insurance-specific factor that is sparsely considered so far. The latter discovery implies that customers prefer a communication channel choice based on the individual circumstances of the claim.
During the Corona-Pandemic, information (e.g. from the analysis of balance sheets and payment behavior) traditionally used for corporate credit risk analysis became less valuable because it represents only past circumstances. Therefore, the use of currently published data from social media platforms, which have shown to contain valuable information regarding the financial stability of companies, should be evaluated. In this data e. g. additional information from disappointed employees or customers can be present. In order to analyze in how far this data can improve the information base for corporate credit risk assessment, Twitter data regarding the ten greatest insolvencies of German companies in 2020 and solvent counterparts is analyzed in this paper. The results from t-tests show, that sentiment before the insolvencies is significantly worse than in the comparison group which is in alignment with previously conducted research endeavors. Furthermore, companies can be classified as prospectively solvent or insolvent with up to 70% accuracy by applying the k-nearest-neighbor algorithm to monthly aggregated sentiment scores. No significant differences in the number of Tweets for both groups can be proven, which is in contrast to findings from studies which were conducted before the Corona-Pandemic. The results can be utilized by practitioners and scientists in order to improve decision support systems in the domain of corporate credit risk analysis. From a scientific point of view, the results show, that the information asymmetry between lenders and borrowers in credit relationships, which are principals and agents according to the principal-agent-theory, can be reduced based on user generated content from social media platforms. In future studies, it should be evaluated in how far the data can be integrated in established processes for credit decision making. Furthermore, additional social media platforms as well as samples of companies should be analyzed. Lastly, the authenticity of user generated contend should be taken into account in order to ensure, that credit decisions rely on truthful information only.
Bluetooth ist ein weit verbreitetes drahtloses Übertragungsprotokoll, das in vielen mobilen Geräten wie bspw. Tablets, Kopfhörer oder Smartwatches verwendet wird. Bluetooth-fähige Geräte senden mehrmals pro Minute öffentliche Advertisements, die u.a. die einzigartige MAC-Adresse des Gerätes beinhalten. Das Mitschneiden dieser Advertisements mittels Bluetooth-Logger ermöglicht es, Bewegungen der Geräte zu analysieren und lassen somit Rückschlüsse auf die Bewegungen der Besitzenden zu.
Zum Schutz der Privatsphäre werden seit 2014 zufällig erzeugte MAC-Adressen in Advertisements verwendet. Eine sog. randomisierte MAC-Adresse bleibt durchschnittlich 15 Minuten lang gültig und wird dann durch eine neue zufällige Adresse ersetzt. Der Aufenthalt eines Geräts zu einem späteren Zeitpunkt kann nicht bestimmt werden. Dennoch kann der Wechsel eines Geräts von einem Bluetooth-Logger zu einem anderen innerhalb dieser 15 Minuten erkannt und somit eine Bewegung des Gerätes abgeleitet werden.
Durch Apps der Kontaktpersonennachverfolgung wie die Corona-Warn-App (CWA) senden auch vermeintlich inaktive Smartphones Bluetooth-Advertisements. Mit etwa einem Viertel der Aufzeichnungen unterstützt die CWA die Auswertungen dieser experimentellen Arbeit.
Um die praktische Anwendbarkeit zu demonstrieren, wurde der Erlebniszoo Hannover als Testgelände genutzt. Die Auswertung der über sieben Wochen gesammelten Daten ermöglichte die Analyse von Stoßzeiten, stark besuchten Orten und Besucherströmen.
In this paper we describe the selection of a modern build automation tool for an industry research partner of ours, namely an insurance company. Build automation has become increasingly important over the years. Today, build automation became one of the central concepts in topics such as cloud native development based on microservices and DevOps. Since more and more products for build automation have entered the market and existing tools have changed their functional scope, there is nowadays a large number of tools on the market that differ greatly in their functional scope. Based on requirements from our partner company, a build server analysis was conducted. This paper presents our analysis requirements, a detailed look at one of the examined tools and a summarizes our comparison of all three tools from our final comparison round.
Unternehmen, die sich ernsthaft mit Nachhaltigkeit beschäftigen, müssen den Nachweis erbringen, dass sie positive Effekte für die Gesellschaft erzielen. Damit ist eine ganzheitliche Wirkungsmessung unabdingbar. Sozialunternehmen sollten als Vorbild für eine solche Wirkungsmessung dienen. Eine wissenschaftliche Studie auf Basis der sog. „Ergebnispyramide“ kommt jedoch zu dem Schluss, dass selbst diese ihre Wirkung bisher kaum ganzheitlich messen.
There are many aspects of code quality, some of which are difficult to capture or to measure. Despite the importance of software quality, there is a lack of commonly accepted measures or indicators for code quality that can be linked to quality attributes. We investigate software developers’ perceptions of source code quality and the practices they recommend to achieve these qualities. We analyze data from semi-structured interviews with 34 professional software developers, programming teachers and students from Europe and the U.S. For the interviews, participants were asked to bring code examples to exemplify what they consider good and bad code, respectively. Readability and structure were used most commonly as defining properties for quality code. Together with documentation, they were also suggested as the most common target properties for quality improvement. When discussing actual code, developers focused on structure, comprehensibility and readability as quality properties. When analyzing relationships between properties, the most commonly talked about target property was comprehensibility. Documentation, structure and readability were named most frequently as source properties to achieve good comprehensibility. Some of the most important source code properties contributing to code quality as perceived by developers lack clear definitions and are difficult to capture. More research is therefore necessary to measure the structure, comprehensibility and readability of code in ways that matter for developers and to relate these measures of code structure, comprehensibility and readability to common software quality attributes.
Mobile crowdsourcing refers to systems where the completion of tasks necessarily requires physical movement of crowdworkers in an on-demand workforce. Evidence suggests that in such systems, tasks often get assigned to crowdworkers who struggle to complete those tasks successfully, resulting in high failure rates and low service quality. A promising solution to ensure higher quality of service is to continuously adapt the assignment and respond to failure-causing events by transferring tasks to better-suited workers who use different routes or vehicles. However, implementing task transfers in mobile crowdsourcing is difficult because workers are autonomous and may reject transfer requests. Moreover, task outcomes are uncertain and need to be predicted. In this paper, we propose different mechanisms to achieve outcome prediction and task coordination in mobile crowdsourcing. First, we analyze different data stream learning approaches for the prediction of task outcomes. Second, based on the suggested prediction model, we propose and evaluate two different approaches for task coordination with different degrees of autonomy: an opportunistic approach for crowdshipping with collaborative, but non-autonomous workers, and a market-based model with autonomous workers for crowdsensing.
Pathologists need to identify abnormal changes in tissue. With the developing digitalization, the used tissue slides are stored digitally. This enables pathologists to annotate the region of interest with the support of software tools. PathoLearn is a web-based learning platform explicitly developed for the teacher-student scenario, where the goal is that students learn to identify potential abnormal changes. Artificial intelligence (AI) and machine learning (ML) have become very important in medicine. Many health sectors already utilize AI and ML. This will only increase in the future, also in the field of pathology. Therefore, it is important to teach students the fundamentals and concepts of AI and ML early in their studies. Additionally, creating and training AI generally requires knowledge of programming and technical details. This thesis evaluates how this boundary can be overcome by comparing existing end-to-end AI platforms and teaching tools for AI. It was shown that a visual programming editor offers a fitting abstraction for creating neural networks without programming. This was extended with real-time collaboration to enable students to work in groups. Additionally, an automatic training feature was implemented, removing the necessity to know technical details about training neural networks.
Music streaming platforms offer music listeners an overwhelming choice of music. Therefore, users of streaming platforms need the support of music recommendation systems to find music that suits their personal taste. Currently, a new class of recommender systems based on knowledge graph embeddings promises to improve the quality of recommendations, in particular to provide diverse and novel recommendations. This paper investigates how knowledge graph embeddings can improve music recommendations. First, it is shown how a collaborative knowledge graph can be derived from open music data sources. Based on this knowledge graph, the music recommender system EARS (knowledge graph Embedding-based Artist Recommender System) is presented in detail, with particular emphasis on recommendation diversity and explainability. Finally, a comprehensive evaluation with real-world data is conducted, comparing of different embeddings and investigating the influence of different types of knowledge.