Scientific publications of the Saarland University
Not a member yet
14321 research outputs found
Sort by
Incorporating knowledge about entities in conversational search
Get PDFAlthough trained on a vast amount of data, language models (LMs) struggle to fully capture information about entities. This issue is especially noticeable for tail entities, which are sparsely covered or entirely absent from knowledge bases. Modern information retrieval (IR) methods rely on language models. Therefore, they struggle to interpret tail entities and questions about them. To bridge this gap, we propose incorporating knowledge about entities into LM-based models in information retrieval. Our first IR method, EVA, addresses the challenge of leveraging knowledge about entities to understand questions. Our second method, CONSENT, tackles the challenges of contextualization and handling informal questions in conversational IR setting. This approach particularly focuses on the challenge of understanding questions about tail entities. We propose our third method, EECATS, to fight three challenges simultaneously: contextualizing questions, handling long-tail entities in conversational IR, and maintaining efficiency for interactive responses. In EECATS, we fine-tune a Sentence-BERT model to infer relatedness between entities in the conversation history and the current question. The current question is then expanded with the most related entities. This expanded question is fed through a retrieval and reranking pipeline. EECATS achieves efficiency through fast score aggregation and effectiveness through fine-tuning the Sentence-BERT model for entity selection.Obwohl Sprachmodelle (Language Models: LMs) mit großen Datenmengen trainiert wurden, haben sie Schwierigkeiten, Informationen über Entitäten zu erfassen. Dieses Problem ist besonders bei Tail-Entitäten spürbar, die in Wissensdatenbanken spärlich oder gar nicht vorhanden sind. Tatsächlich ist das Wissen von Sprachmodellen über Tail-Entitäten begrenzt, da diese im vortrainierten Datensatz nicht häufig vorkommen. Moderne Methoden des Information Retrieval (IR) basieren auf Sprachmodellen. Daher haben sie Schwierigkeiten, Tail-Entitäten und Fragen dazu zu interpretieren. Um diese Lücke zu schließen, schlagen wir vor, Entitätswissen in LM-basierte Modelle für das Information Retrieval zu integrieren. Diese Dissertation präsentiert drei neuartige Methoden zur Verbesserung des aktuellen Stands der Technik: Unsere erste Methode, EVA, befasst sich mit der Herausforderung, Wissen über Entitäten zum Verständnis von Fragen zu nutzen. Dieser Ansatz kombiniert und integriert Wissen aus vortrainierten Wikipedia-Entitäteinbettungen in die dichte Repräsentation des Sprachmodells BERT. Wir verwenden einen neuartigen Ansatz, um mehrere Darstellungsansichten für jedes Dokument vorzuberechnen und so die Wahrscheinlichkeit zu maximieren, Benutzerabsichten zu erfassen. Jede Ansicht kodiert einen bestimmten Aspekt basierend auf einer Reihe verwandter Entitäten innerhalb eines Dokuments. Unsere zweite Methode, CONSENT, befasst sich mit den Herausforderungen der Kontextualisierung und Verarbeitung informeller Fragen in Konversationen. Dieser Ansatz konzentriert sich auf das Verständnis von Fragen zu Long-Tail-Entitäten. Wir entwickeln CONSENT, um die aktuelle Frage mit einer ausgewählten Teilmenge vorheriger Dialogschritte und Entitäten zu kontextualisieren. In dem CONSENT verwenden wir Sentence-BERT in einem Zero-Shot-Setting, um Ähnlichkeiten zwischen Entitäten, Dialogschritten, der aktuellen Frage und Antwortoptionen zu ermitteln. Basierend auf diesen Ähnlichkeiten werden Entitätsauswahl und Antwortrangfolge gemeinsam modelliert und mithilfe der ganzzahligen linearen Programmierung gelöst. Mit unserer dritten Methode, EECATS, bewältigen wir drei Herausforderungen: die Kontextualisierung von Fragen, den Umgang mit den Long-Tail-Entitäten in Konversationen und die Aufrechterhaltung der Effizienz interaktiver Antworten. Wir verwenden Fine-Tuning für ein Sentence-BERT-Modell, um die Verwandtschaft zwischen Entitäten im Gespräch und der aktuellen Frage zu bestimmen in EECATS. Die aktuelle Frage wird dann um die am engsten verwandten Entitäten erweitert. Diese erweiterte Frage wird durch eine Retrieval- und Ranking-Pipeline gespeist. EECATS erreicht Effizienz durch schnelle Score-Aggregation und Effektivität durch die Feinabstimmung des Sentence-BERT-Modells für die Entitätsauswahl
Probing Rate-Dependent Liquid Shear Viscosity Using Combined Machine Learning and Nonequilibrium Molecular Dynamics
Get PDFAccurately measuring liquid dynamic viscosity across a wide range of shear rates, from the linear-response to shear-thinning regimes, presents significant experimental challenges due to limitations in resolving high shear rates and controlling thermal effects. In this study, we integrated machine learning (ML) with nonequilibrium molecular dynamics (NEMD) simulations to address these challenges. A supervised artificial neural network (ANN) model was developed to predict viscosity as a function of shear rate, normal pressure, and temperature, effectively capturing the complex interplay among these variables. The model reveals distinct trends in shear viscosity, characterized by the shear-thinning exponent, and highlights nonmonotonic behavior in the radius of gyration components, reflecting molecular morphological changes driven by rate-dependent volume expansion. Notably, temperature effects diminish at higher shear rates, where molecular alignment and spacing dominate the response to shear. By implementing the ‘fix npt/sllod’ command in LAMMPS, we achieve precise constant-pressure control in NEMD simulations, ensuring accurate representation of system dynamics. This study demonstrates the potential of ML-enhanced NEMD for efficient and accurate viscosity prediction, providing a robust framework for future research in complex fluid dynamics and material design
Lifetime Variations in Prolactin Expression in the Hippocampus and Dentate Gyrus of the Rat
Get PDFProlactin is a hormone with demonstrated roles in the brain, including neurogenesis,
neuroprotection, learning, stress response or memory consolidation. To determine the
prolactin expression in the rat hippocampus during aging and to resolve some controversies
related to the presence of prolactin in the hippocampus, the aim of this study was to analyze
whether changes occur in the expression of prolactin during different stages of life. To
determine this, we designed an experimental study in which we analyzed the expression
and location of prolactin in the rat hippocampus, Ammon’s horn and Dentate Gyrus, during
different stages of life (prepubertal, postpubertal, young adult, adult and old) and checked
if there are differences related to sex. Overall, the results obtained show that prolactin is
present in the rat hippocampus and that prolactin is synthesized, as deduced from the
findings obtained via ELISA, immunohistochemistry, qPCR and in situ hybridization. After
analyzing the correlation between serum and hippocampal prolactin levels and comparing
the amounts of Prl mRNA and the hormone, the results obtained suggest that hippocampal
prolactin has a dual origin: local synthesis of the hormone and its passage from the blood.
On the other hand, the amounts of prolactin and its mRNA in the hippocampus vary
with sex and age, suggesting the existence of age-related sexual dimorphism. The results
obtained suggest that hippocampal aging is related to a decrease in the hippocampal
prolactin system, which helps to better understand brain aging
Optimizing superparamagnetic ferrite nanoparticles: microwave-assisted vs. thermal decomposition synthesis methods
No full textSuperparamagnetic iron oxide nanoparticles are of crucial importance for various applications in medicine and biology as well as in materials science, where properties such as magnetism and inductive heating are advantageous. In this study, we systematically compare the synthesis methods for ferrite nanoparticles with those of pure iron oxide, focusing on their final properties. We synthesized superparamagnetic substituted ferrite nanoparticles with an average diameter of 5 to 8 nm with the general formula of MxFe3−xO4 (M = Fe2+, Mn2+, Co2+) using both conventional thermal decomposition (TD) method and microwave-assisted (MW) methods. Although the manganese-substituted particles obtained through both methods exhibited a narrow size distribution and high surface coverage with oleic acid, they demonstrated lower heating efficiency in an induction field compared to the cobalt-substituted particles. In particular, the replacement of Fe2+ ions with Co2+ ions significantly improved the self-heating ability and increased the specific absorption rate (SAR) from 22.7 for Fe3O4 to 106.3 W gNP−1 for Co0.88Fe2.12O4 nanoparticles. In addition, the concentration of 1,2-dodecanediol in the reaction mixture significantly influenced the shape and size distribution of the particles. Microwave-assisted synthesis resulted in higher incorporation of M2+ ions, as confirmed by ICP-MS and EDX spectroscopy, and more uniform particle sizes due to homogeneous nucleation. By optimizing the microwave method, we were able to produce small size superparamagnetic particles with high saturation magnetization (89.2 emu g−1 at 300 K), capable of generating more heat in the magnetic field, making these particles suitable candidates for induction heating in materials
In situ real-time monitoring of ammonium, potassium, chloride and nitrate in small and medium-sized rivers using ion-selective-electrodes – a case study of feasibility
No full textReal-time measurements are particularly important for monitoring especially small and medium-sized highly dynamic rivers. Their results are indispensable for planning (cost)efficient measures to improve the chemical and ecological quality of the rivers. It is therefore expedient to be able to use cost-effective, reliable equipment. In theory, ion-selective electrodes (ISEs) are excellently suited for this purpose due to their cost-effectiveness, their high spatial flexibility and their self-sufficient energy supply for the simultaneous recording of different parameters. In practice, however, malfunctions occur caused by temperature changes, interferences of non-target ions or long-term-drifts. This study investigated the applicability of ISEs for in situ real-time measurement of ammonium (NH4+), potassium (K+), chloride (Cl−) and nitrate (NO3−) in small rivers. ISEs from three different manufacturers were deployed for five months in a river monitoring station. The measured data were compared with real-time data gained from an on-line photometer, a gas-sensitive analyser and an optical UV probe as well as with grab samples analysed using ion chromatography (IC). Special attention was given to the challenges posed by low concentrations, temperature fluctuations, concentration changes of the analytes, and interfering ions, as well as to the long-term stability and the lifetime of the ISEs. For Cl− and NO3− good agreements with the comparative measurements were found, and it could be shown that the ISEs are well suited for event detection of all four observed parameters. Temperature compensation is the main challenge that complicates application in highly dynamic rivers. Therefore, further optimisation is urgently needed for reliable quantitative analysis, which is part of future work
Shaking up the foundations of modern separation logic
Get PDFThe problem of how to scalably verify large, stateful programs is one of the oldest—and still unsolved—challenges of computer science. Over the last two decades, there has been considerable progress toward this goal with the advent of separation logic, a verification technique for modularly reasoning about stateful programs. While originally only developed for imperative, pointer-manipulating programs, separation logic has in its modern form become an essential tool in the toolbox of the working semanticist for modeling programming languages and verifying programs. This dissertation presents a line of work that revisits the foundations of modern separation logic in the context of the separation logic framework Iris. It targets two broader areas: step-indexing and automation. Step-indexing is a powerful technique for modeling many of the advanced, cyclic features of modern languages. Here, Transfinite Iris shows how to generalize step-indexing from proving safety properties to proving liveness properties, and Later Credits enable more flexible proof patterns for step-indexing based on separation logic resources. Automation is important for reducing the overhead of verification to scale to larger code bases. Here, Quiver introduces a new form of guided specification inference to reduce the specification overhead of separation logic verification, and Daenerys develops new resources in Iris that lay the groundwork for automating parts of Iris proofs using SMT solvers.Wie man skalierbar große Programme verifiziert ist eine der ältesten, ungeklärten Fragen der Informatik. In den letzten zwei Jahrzehnten wurde hier deutlicher Fortschritt erzielt mit der Einführung von Separationslogik, einer Technik für modulare Programmverifikation. Separationslogik wurde ursprünglich nur für imperative Programme mit Zeigern entwickelt, ist aber in ihrer modernen Form essenziell geworden, um Programmiersprachen zu modellieren und Programme zu verifizieren. Diese Dissertation präsentiert eine Reihe von Arbeiten, die sich im Kontext der Separationslogik Iris mit den Grundlagen moderner Separationslogik beschäftigen. Sie fokussiert sich auf zwei Bereiche: Step-Indexing und Automatisierung. Step-Indexing ist eine wichtige Modellierungstechnik für viele der fortgeschrittenen, zyklischen Funktionen moderner Programmiersprachen. Hier zeigt Transfinite Iris, wie man Step-Indexing nicht nur für Sicherheitseigenschaften verwenden kann, sondern auch um Lebendigkeitseigenschaften zu beweisen, und Later Credits ermöglichen flexiblere Beweisstrukturen durch die Verwendungen von Separationslogikressourcen. Automatisierung ist wichtig, um den Aufwand von Verifikation zu senken, um größere Programme zu verifizieren. Hier entwickelt Quiver eine neue Form der Spezifikationsinferenz, um den Aufwand von Separationslogikspezifikationen zu reduzieren, und Daenerys entwickelt neue Iris-Ressourcen, die den Weg bereiten, um Teile von Iris-Beweisen mit SMT-Solvern zu automatisieren.This research was supported by a Google PhD Fellowship
Pädiatrische Subspezialisierungen in Deutschland im Jahr 2024: Eine kritische Bestandsaufnahme
Get PDFHintergrund: Über die spezialisierte Versorgung von Kindern und Jugendlichen im stationären Bereich durch entsprechend weitergebildete Kinder- und Jugendmediziner gibt es in Deutschland bisher kaum Daten. Diese Arbeit ist eine aktuelle Bestandsaufnahme zu pädiatrischen Subspezialisierungen.
Methodik: Mittels drei verschiedener Datenerfassungen wurden Daten zu der Anzahl und Verteilung der Subspezialisierungen erhoben: Fragebögen, die an deutsche Kinderkliniken und Klinikabteilungen gesendet wurden, Recherche von Daten über die Websites der Kliniken und anderen öffentlich zugänglichen Quellen sowie Erfassung der Daten der Bundesärztekammer, der Landesärztekammern und des Statistischen Bundesamtes. Es werden Grundlagen der Gesundheitsversorgung in Deutschland im Hinblick auf die Pädiatrie beschrieben und die Daten zu den Subspezialisierungen nach unterschiedlichen Kriterien, beispielsweise Position, Geschlecht, Verteilung Ost-/Westdeutschland, analysiert.
Ergebnisse: Die Neonatologie ist mit ungefähr einem Viertel aller Subspezialisierung die mit Abstand häufigste, gefolgt von der Intensivmedizin, der Neuropädiatrie und der Kinder- und Jugend-Kardiologie. Die am seltensten repräsentierten Subspezialisierungen sind die Immunologie und Hämostaseologie. Pro 100.000 Einwohner unter 18 Jahren stehen in Deutschland circa sieben Neonatologen, zwei bis drei Intensivmediziner, Neuropädiater und Kinder- und Jugendkardiologen zur Verfügung, aber nur 0,2 Immunologen und Hämostaseologen.
Bei der Geschlechterverteilung zeigt sich, dass der Frauenanteil der Kinder- und Jugendmediziner durch den Erwerb einer Subspezialisierung und mit steigender Position deutlich abnimmt. Fast drei Viertel aller subspezialisierten Fachärzte sind weiblich, doch weniger als die Hälfte der Oberarztpositionen sowie nur jede fünfte Chefarztposition ist von Frauen besetzt. Es zeigt sich eine hohe positive Korrelation zwischen der Anzahl der Subspezialisierungen und der Anzahl der Weiterbildungsbefugnisse. In Ostdeutschland ist der Anteil der Einwohner unter 18 Jahren geringer als in Westdeutschland, es stehen jedoch pro Einwohner unter 18 Jahren mehr Kinder- und Jugendmediziner, mehr Subspezialisierungen und mehr Weiterbildungsbefugnisse zur Verfügung.
Schlussfolgerung: Für eine flächendeckende patientengerechte Versorgung von Kindern und Jugendlichen wird eine bedarfsgerechte Anzahl und Verteilung von subspezialisierten Kinder- und Jugendmedizinern benötigt. Da sich dieser Bedarf stetig ändert, ist die kontinuierliche Koordination von Fachärzten, subspezialisierten Fachärzten und Weiterbildungsbefugnissen anhand aktueller Daten erforderlich.
Stichwörter: Kinder- und Jugendmedizin, Versorgung, Subspezialisierung, Schwerpunktbezeichnung, Zusatzbezeichnung, Weiterbildung, Geschlechterverteilung, Ost-/West-VerteilungBackground: Currently there is little data available concerning subspecialized pediatricians, responsible for the specialized in-patient care of children and adolescents in Germany.
This study assesses the number of subspecialized pediatricians working in the inpatient care in Germany. Methods: The collection of data was conducted through three different methods: A survey sent to German children’s hospitals and pediatric departments, a research of publicly available data from hospital websites and other publicly accessible websites, as well as a collection of data through German health institutions (i.e.the German Medical Association (Bundesärztekammer) and the state chambers of physicians (Landesärztekammern) of the different German states, as well as the German Federal Statistical Office (Statistisches Bundesamt)). The study describes the pediatric health care system in Germany and the collected data on subspecialties is analyzed based on criteria such as position, sex, and regional distribution in East and West Germany.
Results: The subspeciality neonatology accounts for around three quarters of all subspecialities and therefore is the most common subspeciality in pediatrics, followed by intensive care medicine, neuropediatrics and pediatric cardiology. The least common subspecialities are immunology and hemostaseology. Per 100.000 inhabitants under the age of 18 there are around seven neonatologists, two to three intensive care physicians, neuropediatricians and pediatric cardiologists, but only 0,2 immunologists and hemostaseologists.
The sex distribution shows a high proportion of female pediatricians which decreases with the acquisition of subspecialities and decreases even further with increasing hierarchy level. Nearly three quarters of subspecialized pediatricians are women, however, less than half of the senior physician positions, and only every fifth chief physician position is filled by a woman. The study shows a positive correlation between the number of subspecialists and the respective training positions. Although Eastern Germany has a lower proportion of inhabitants under the age of 18, it has more pediatricians, subspecializations and training positions per 100.000 inhabitants under the age of 18 than Western Germany.
Conclusions: The appropriate number and distribution of subspecialized pediatricians, adapted to the needs, is necessary to ensure adequate patient-oriented health care of children and adolescents. Since the demand is constantly changing, it is essential to continuously coordinate the number and distribution of pediatricians and subspecialized pediatricians as well as pediatric training positions by means of the current data.
Keywords: pediatric and adolescent medicine, pediatric health care, subspecialty, subspecialty training, gender distribution, East/West distributio
Additive manufacturing of Ni62Nb38 metallic glass via laser powder bed fusion
Get PDFThe binary Ni62Nb38 bulk metallic glass exhibits attractive mechanical properties like a yield strength of more than 3 GPa and a hardness of more than 900 Vickers. Due to its limited glass forming ability in casting processes, industrial applications of this impressive material are still pending. Additive manufacturing via laser powder bed fusion (PBF-LB/M) can allow to overcome these limitations. Therefore, we present the first PBF-LB/M parameter study on Ni62Nb38. We are able to achieve high densification and almost fully amorphous samples, thereby demonstrating the general applicability of PBF-LB/M to process binary amorphous Ni–Nb alloys. Furthermore, two challenges can be identified, which have to be addressed before additively formed Ni62Nb38 can be considered for commercial use. Future parameter fine-tuning must lead to fully vitrified samples, and one must also find a way to avoid crack formation during processing, which was found to be one of the main issues in the present study
Against the words : The importance of non-verbalities for the negotiation of self-determination in open experimentation
Get PDFDie Studie doing AGENCY untersucht am Beispiel des Offe nen Experimentierens, wie Selbstbestimmung zwischen Schülerinnen und Schülern
und ihrer Lehrperson ausgehandelt wird. Um die Interaktionen von Schülerinnen
und Schülern und ihrer Lehrperson beim Offenen Experimentieren zu beobachten,
kontextualisiert zu analysieren sowie hinsichtlich der Aushandlungsprozesse zwi schen den Schülerinnen und Schülern und ihrer Lehrperson sensibel interpretieren
zu können, wurde die Grounded Theory Methodologie als qualitativer Forschungs zugang gewählt. Die Datengrundlage bildeten Beobachtungsprotokolle von offenen
Experimentiersituationen, die nach den Verfahren der Grounded Theory Methodo logie iterativ und mehrstufig kodiert wurden, bis eine Grounded Theory mittels
„gesättigtem“ Kategoriensystem entwickelt werden konnte.
Ein Teilergebnis dieser Grounded Theory weist auf den massiven Einfluss der
Kommunikation der Lehrperson – und besonders im Fall einer starken Diskrepanz
zwischen Verbalitäten und Nonverbalitäten – auf die Experimentierhandlungen der
Schülerinnen und Schüler hin. Insbesondere die nonverbale Kommunikation der
Lehrperson wirkt sich dabei deutlich, ggf. auch „wider die Worte“, auf das Vorgehen
der Schülerinnen und Schüler aus.The study doing AGENCY uses the example of open experimentation to
investigate how self-determination is negotiated between pupils and their teacher.
Grounded theory methodology was chosen as a qualitative research approach. Hence
it was possible to observe the interactions between pupils and their teacher during
open experimentation, to analyse them in a contextualised manner and to interpret
them sensitively regarding the negotiation processes between the pupils and the
teacher. The data basis was formed by observation protocols of open experimen tation situations, which were coded iteratively and, in several stages, according to
the procedures of grounded theory methodology until a grounded theory could be
developed using a ‘saturated’ category system. A partial result of this grounded the ory points to the massive influence of the teacher’s communication—and especially
in the case of a strong discrepancy between the teacher’s verbalisations and non verbalisations—on the students’ experimental actions. In particular, the teacher’s
non-verbal communication has a clear effect on the students’ actions, possibly even
“against the words”
Deciphering the performance of different surface models for corneal topography
Get PDFPurpose
To study the performance of different corneal surface models to be used for ray tracing. Models based on geometric surfaces and polynomial fits were compared and the differences discussed.
Methods
For this simulation study, five characteristic generic surface configurations were generated: (A) perfect biconic, (B) decentred biconic with white noise, (C) biconic with paracentral hollow simulating the situation after myopic LASIK, (D) biconic with random dot irregularities and (E) rotationally symmetric conic with mid-peripheral bump simulating the situation of corneal ectasia. A floating best fit sphere (BFS), conic (BFC), biconic (BFBC), fringe Zernike on top of a BFS (BFSZ), fringe Zernike (BFZ) and Gaussian process surface model (BFGP) were fitted and the root-mean-squared fit error was analysed.
Results
Surfaces A and B were well described by BFBC, BFSZ, BFZ and BFGP, but not by BFS and BFC. Surface C was not well represented by BFS, BFC and BFBC, but reasonably with BFSZ and BFZ and quite well with BFGP. Surfaces D and E were poorly represented, especially with BFS, BFC and BFBC, but also with BFSZ and BFZ and quite well with BFGP. There was no systematic difference between the two Zernike representations BFSZ and BFZ, even for surface B.
Conclusions
Representing corneal point cloud data with a closed surface model plays a key role in ray tracing. Simple surface models such as BFS, BFC or BFBC are easy to handle but do not fully represent clinical situations with local irregularities after corneal refractive surgery or with ectasia