http://hdl.handle.net/10900/42133 Tue, 19 May 2026 00:24:24 GMT 2026-05-19T00:24:24Z http://hdl.handle.net/10900/172841 The Impact of AI-based Writing Recommendations for Emotional Tone on Interpersonal Online Interactions Hagedorn, Josephine Camille Die Dissertation ist gesperrt bis zum 07. September 2027 ! Tue, 07 Sep 2027 00:00:00 GMT http://hdl.handle.net/10900/172841 2027-09-07T00:00:00Z http://hdl.handle.net/10900/172811 Detecting Ice Fabric and Englacial Structures with Phase-Sensitive Radar Oraschewski, Falk Marius Gebirgsgletscher und polare Eisschilde verlieren als Reaktion auf die globale Erwärmung in zunehmender Geschwindigkeit an Masse. Der Massenzuwachs durch Schneeakkumulation und der Masseverlust durch Kalbung und Schmelzen werden in erster Linie durch klimatische Faktoren gesteuert. Die Fließdynamik des Eises bestimmt jedoch, wie schnell die Masse von den Akkumulations- zu den Ablationszonen transportiert wird. Aus diesem Grund ist die Eisdynamik ein wichtiger Einflussfaktor für die Stabilität und Entwicklung von Eisschilden und Gletschern. Daher sind genaue Parametrisierungen der eisdynamischen Prozesse für zuverlässige Vorhersagen über den künftigen Massenverlust und den Anstieg des Meeresspiegels unerlässlich. Die Eisdynamik prägt darüber hinaus die Eiskristallstruktur, welche die Orientierungsverteilung der Eiskristalle beschreibt, und englazialen Strukturen, wie die stratigraphische Schichtung des Eises. Diese Strukturen können durch Radare erfasst werden, da Eis eine hohe Transparenz für Radiowellen aufweist. Folglich kann aus Radardaten auf vergangene Eisverformungen geschlossen werden, um dynamische Prozesse im Eis zu untersuchen. Diese Arbeit präsentiert zwei Studien, in denen neue Erfassungsmethoden für ein phasensensitives Radar (pRES) entwickelt werden, um die tiefe englaziale Stratigraphie von Gebirgsgletschern und die vertikale Anisotropie der Eiskristallstruktur zu bestimmen. In der ersten Studie wird das pRES mobilisiert, um ein phasensensitives Radarsystem zu erhalten, das für den Einsatz auf Gebirgsgletschern geeignet ist. Eine Proof-of-Concept-Studie, die am Colle Gnifetti, Schweiz und Italien, durchgeführt wurde, zeigt, dass das System die tiefe englaziale Stratigraphie abbilden kann, die von zuvor eingesetzten Impulsradaren nicht erfasst werden konnte. Ein schichtoptimiertes Verarbeitungsschema für das Radar mit synthetischer Apertur wird an die mobilen pRES-Daten angepasst, um Störsignale zu unterdrücken und die Klarheit der Reflexionshorizonte zu erhöhen. In Kombination ermöglichen diese Techniken die Erkennung der englazialen Stratigraphie über nahezu die gesamte Eisdicke. Diese Verbesserung wird durch den ältesten Reflexionshorizont veranschaulicht, der in den gewonnenen Daten durchgängig nachweisbar ist und ein Alter von 288 ± 35 a aufweist, verglichen mit 78 ± 12 a, das in früheren Untersuchungen erreicht wurde. Allerdings konnten die zuvor vorgeschlagenen eisdynamisch geformten komplexen englazialen Strukturen, wie Schichtstörungen und englaziale Falten, nicht nachgewiesen werden. Stattdessen wird die Veränderung der Reflektivität auf eine erhöhte Ablagerungsrate säurehaltiger Verunreinigungen im oberen Eis zurückgeführt, die mit dem Beginn der Industrialisierung korreliert. Die zweite Studie behandelt die Erkennung vertikaler Kristallanisotropie durch phasenbasierte polarimetrische Weitwinkelradarmessungen. Bisherige Polarimetriestudien verwendeten typischerweise Nadir-ausgerichtete Antennenkonfigurationen, die in erster Linie für horizontale Kristallanisotropie sensitiv sind. Diese Studien setzen die Annahme voraus, dass die Eiskristallstruktur vertikal ausgerichtet ist, eine Annahme, die in komplexen Eisflussregimen möglicherweise nicht gilt. Frühere theoretische Studien haben vorgeschlagen, dass die vertikale Anisotropie der Eiskristallstruktur und deren Neigung mittels polarimetrischer Weitwinkelmessungen mit schrägen Einfallswinkeln der Radarwellen bestimmt werden können. Um diesen Ansatz zu testen, wurde auf dem Ekström-Schelfeis, Ostantarktis, eine polarimetrische Common-Midpoint-Radarmessung durchgeführt. Die vorgestellte Untersuchungsmethode ist sowohl für die vertikale als auch die horizontale Kristallanisotropie sensitiv. Diese wird über Phasenverschiebungen zwischen unterschiedlich polarisierten Radarwellen erreicht, die durch die Anisotropie der dielektrischen Permittivität verursacht werden. Es wird ein Inversionsverfahren eingesetzt, um alle relevanten Komponenten der Eiskristallstruktur abzuleiten. Dabei zeigt sich, dass die Kristallanisotropie die beobachteten Phasenunterschiede nicht alleine erklären kann. Die strukturelle Firnanisotropie wird als zusätzliche Quelle dielektrischer Anisotropie identifiziert, welche die Signaturen der Eiskristallstruktur überlagert. Da die Firnstruktur in erster Linie die vertikale dielektrische Anisotropie beeinflusst, wurde sie in früheren Polarimetriestudien mit Nadir-ausgerichteten Antennen vernachlässigt. Durch den Nachweis, dass Radar die Firnanisotropie detektieren kann, legt diese Studie nahe, dass gezielte Radarpolarimetriemessungen zu einem besseren Verständnis der strukturellen Firnanisotropie beitragen und Kalibrierungsdaten für die Satellitenfernerkundung von Firn liefern können. Zusammenfassend lässt sich sagen, dass die vorgestellten Studien den Anwendungsbereich von phasensensitiven und polarimetrischen Radaren erweitern und neue Perspektiven für die Interpretation phasensensitiver Radardaten bieten. Beide Studien stellen jedoch nur erste Proof-of-Concepts für diese neuen Erfassungstechniken dar und ebnen den Weg für ihre Anwendung bei der Untersuchung komplexer dynamischer Prozesse im Eis in der zukünftigen Forschung. Um diese Bemühungen zu fördern, regt diese Arbeit die Entwicklung neuer Radarsysteme an, die speziell für die Erkennung der neu identifizierten Radarziele ausgelegt sind.; Mountain glaciers and polar ice sheets are losing mass at accelerating rates in response to global warming. The rates of mass gain by snow accumulation and mass loss by calving and melting processes are primarily controlled by climatic factors. However, the flow dynamics of ice determine how fast mass is transported from accumulation to ablation zones. For this reason, ice dynamics are an important control on the stability and evolution of ice sheets and glaciers. Therefore, accurate parametrizations of ice-dynamical processes are essential for reliable projections of future mass loss and sea level rise. Ice dynamics imprint on ice fabric, which describes the orientation distribution of ice crystals, and on englacial structures, such as the stratigraphic layering of ice. These englacial features can be detected by radar as ice is highly transparent to radio waves. Consequently, past ice deformation can be inferred from radar data to study ice dynamical processes. This thesis presents two studies in which new acquisition methods for the phase-sensitive Radio Echo Sounder (pRES) are developed to detect the deep englacial stratigraphy of mountain glaciers and vertical ice fabric anisotropy. In the first study, the pRES is mobilized to establish a phase-sensitive radar system suitable for deployment on mountain glaciers. A proof-of-concept survey conducted at Colle Gnifetti, Switzerland and Italy, demonstrates that the system can resolve the deep englacial stratigraphy, which could not be detected by previously deployed impulse radars. A layer-optimized synthetic aperture radar processing scheme is adapted to the mobile pRES data to suppress clutter and enhance the clarity of specular reflection horizons. In combination, these techniques enable the detection of englacial stratigraphy throughout essentially the entire ice thickness. This improvement is exemplified by the oldest reflection horizon that is continuously traceable in the acquired data, which has an age of 288 ± 35 a, compared to 78 ± 12 a, achieved in previous surveys. However, previously suggested ice-dynamically induced complex englacial structures, such as layer disturbances and englacial folds, could not be detected. Instead, the change in reflectivity is attributed to an increased deposition rate of acidic impurity layers in the upper ice, correlating with the onset of industrialization. The second study focuses on detecting vertical ice fabric anisotropy through phase-based polarimetric wide-angle radar observations. Previous polarimetry studies typically employed nadir-looking antenna configurations, which are primarily sensitive to horizontal ice fabric anisotropy. These studies rely on the assumption that ice fabric is vertically aligned, a presumption that may not hold in complex flow regimes. Previous theoretical studies suggest that vertical ice fabric anisotropy and tilted ice fabric patterns can be inferred from polarimetric wide-angle radar surveys that feature oblique radio wave propagation. To test this approach, a polarimetric common midpoint radar survey was conducted on Ekström Ice Shelf, East Antarctica. The presented survey is sensitive to both vertical and horizontal ice fabric anisotropy, utilizing phase shifts between differently polarized radar waves caused by the anisotropy of the dielectric permittivity. An inversion framework is implemented to infer all relevant ice fabric components, indicating that ice fabric alone cannot explain the observed phase differences. Structural firn anisotropy is identified as an additional source of dielectric anisotropy that superimposes the ice fabric signatures. As the firn structure primarily impacts the vertical dielectric anisotropy, it has been neglected in previous nadir-looking polarimetry studies. By demonstrating that radar can measure firn anisotropy, this study suggests that targeted radar polarimetry can contribute to an improved understanding of structural firn anisotropy and provide ground-truth data for satellite remote sensing of firn. In synthesis, the presented studies expand the application range of phase-sensitive and polarimetric radar and provide new perspectives for interpreting phase-sensitive radar data. However, both studies only represent first proof-of-concepts for these new acquisition techniques, paving the way for their broader application in future research to investigate complex ice dynamical processes. To foster these efforts, this thesis encourages the development of new radar systems that are specifically designed for detecting the newly identified radar targets. Thu, 04 Dec 2025 00:00:00 GMT http://hdl.handle.net/10900/172811 2025-12-04T00:00:00Z http://hdl.handle.net/10900/172808 Do we really remember the truth? Factors that influence the success and failure of veracity retrieval Antes, Nicole People navigate a world shaped by a constant flood of information that does not always draw clear boundaries between what is true and what is false. To navigate such blended information environments, people must remember not only the content of information they encounter but also whether it was true or false and memorize its veracity. This dissertation focuses precisely on such a blended information environment in the context of event descriptions, where both true and false information co-occur. Specifically, it addresses the research question of when people succeed or fail to retrieve information accurately within this context, where some information receives additional veracity information. Attention is given especially to the role of narrative consistency, perceptual veracity cues, modality congruency, and confidence. Across four empirical chapters, I examined four central questions. (1) How do inconsistencies between narratives shape reasoning about the described events? (2) Can confidence serve as a metacognitive signal that helps in navigating inconsistent content? (3) How do perceptual veracity cues influence the accuracy and confidence of veracity retrieval? And (4) How does modality (in)congruency influence the accuracy and confidence of veracity retrieval? After introducing the general theoretical framework in Chapter 1, the first empirical chapter (Chapter 2) addresses how competing but unlabeled causes affect event reasoning. My results showed that both proactive and retroactive interference can occur, suggesting that the persistence of information depends less on the presentation order than on the coexistence of competing causes in memory. Providing a veracity cue as to which cause is accurate, however, can shift the reasoning in the right direction. Chapter 3 extends these findings by incorporating confidence judgments, showing that confidence does not necessarily track accuracy but moderates which cause is later used in reasoning. High confidence was often related to whichever cause was more fluently retrieved, irrespective of its accuracy. Providing a cue to the accurate cause also shifted confidence towards that cause. Thus, confidence was shown not to track accuracy reliably but to act as a metacognitive signal that determined which information was maintained or successfully updated. Chapter 4 investigates perceptual veracity cues embedded in coherent event descriptions. While the presence of such cues reduced overall accuracy, they also revealed asymmetries. False-labeled details were more likely to be forgotten or misremembered as true, and these errors were often related to high confidence. Thus, cues designed to highlight veracity may foster false confidence if the cue is forgotten. Chapter 5 addresses modality inconsistency between encoding and retrieval. Across three experiments, cross-modal retrieval was found to impair veracity retrieval, especially when false-labeled information was encoded in text and later retrieved in graphics. By contrast, information encoded in graphics was more robust to modality shifts, highlighting asymmetries in how encoding formats determine the retrieval of veracity information. Taken together, the chapters in this dissertation show that veracity retrieval is a dynamic process shaped by fluency, cues, modality, and confidence. False-labeled information is particularly fragile, being more likely to be forgotten, misremembered, or confidently misclassified as true. These results extend models of situation model construction, source monitoring, and truth tagging by emphasizing the dynamic role of confidence as a metacognitive factor in veracity retrieval. From an applied perspective, they caution against simplistic interventions such as labeling false information, which can unintentionally increase misplaced confidence. It underlines the need for approaches that promote accurate confidence calibration and resilience across modalities in order to better protect people against misinformation.; Die Dissertation ist gesperrt bis zum 22. Oktober 2026 ! Thu, 22 Oct 2026 00:00:00 GMT http://hdl.handle.net/10900/172808 2026-10-22T00:00:00Z http://hdl.handle.net/10900/172807 Biomedical Machine Learning Beyond the Training Distribution Visonà, Giovanni Machine learning (ML) holds the potential to impact many aspects of our lives, particularly in high-stakes areas like law, autonomous systems, and healthcare. The prospects of leveraging large quantities of data to mine patterns, improve decision-making, and navigate the complexity of biological systems are especially appealing and can have far-ranging consequences; however, ensuring the robustness and reliability of machine learning models has proven a remarkably difficult challenge, leading to considerable efforts by the research community. In particular, understanding how ML models generalize to new observations is a necessary condition for the fruitful translation of these advancements in machine learning to clinical practice or to expand biological domain knowledge. When the training and test settings correspond, and the individual observations do not affect each other---the so-called independent, identically distributed (IID) setting---machine learning and deep learning have displayed remarkable capabilities. But when the data-generating distribution shifts, or when we want to solve related but slightly different tasks, then the quality of the predictions of a model can rapidly deteriorate. In this thesis, I will examine the challenges that arise when generalizing beyond the training distribution in biomedical machine learning and the approaches developed to tackle such challenges. The first part of the thesis will provide a broad overview of the topic of generalization in machine learning, starting from a conceptual formulation of the generalization problem and the progress made in laying theoretical foundations for generalization in ML. Delving into the topic, I will provide an examination of the most common paradigms developed to improve predictive performance when generalizing outside the training distribution, and I will discuss the role of causal reasoning within this picture. Afterwards, I will review the state of biomedical applications of machine learning, highlighting some of the most well-studied areas of research, as well as fields where the use of ML has yet to deliver on its promise. Of particular interest is the topic of biases in biomedical data: given the staggering complexity of biological phenomena, and the considerable experimental constraints on gathering relevant data, it is crucial that we understand how to separate noise and natural variability from meaningful signal. Related to this idea, I will also discuss the ever-present challenge of validating the results of biomedical ML models. Following these broad overviews of generalization and biomedical machine learning, I will present two works revolving around the application of deep learning to biological and clinical data. In each of them, the generalization challenges and paradigms presented in the earlier chapters play a crucial role, enabling novel prediction tasks or revealing insights into the properties of the models. The first work, that focuses on the task of imputing epigenomic signals, showcases how the use of transfer learning enables the out-of-distribution imputation of individual-specific epigenomic patterns, a case study in personalized epigenomics that is, to the best of my knowledge, the first of its kind. Afterwards, I will present a research project that tackles the task of predicting antimicrobial resistance from clinical proteomics data; when delving into the workings of the models proposed, the analysis of zero-shot prediction tasks offers a window into their robustness, which can guide future developments and offer insights for the data collection efforts required to progress further. Wed, 03 Dec 2025 00:00:00 GMT http://hdl.handle.net/10900/172807 2025-12-03T00:00:00Z