31325 research outputs found
Sort by
Leveraging Generalist and Foundation Models for Effective Biomedical Image Segmentation
No full textBiomedizinisches Deep Learning wird oft durch Datenknappheit und Domänenverschiebungen eingeschränkt. Diese Monographie untersucht die Nutzung vortrainierter Modelle, um diese Herausforderungen anzugehen, und präsentiert drei Beiträge. Zuerst demonstriert die PEERCE-Pipeline den praktischen Wert dieses Ansatzes durch die Kombination feinabgestimmter Netzwerke mit einem Generalistenmodell (Cellpose) zur Vorhersage des PD-L1-Scores beim Angiosarkom. Obwohl erfolgreich, verdeutlicht PEERCE die Notwendigkeit sorgfältiger Vorverarbeitung für Cellpose und erfordert weiterhin annotierte Daten für andere Komponenten. Diese Limitation motiviert den zweiten Beitrag, SelfAdapt, ein Framework für unüberwachte Domänenadaption. SelfAdapt ermöglicht die annotationsfreie Anpassung von Generalistenmodellen wie Cellpose an neue Zieldomänen mittels Student-Teacher-Ansatz, L2-SP-Regularisierung und annotationsfreien Abbruchkriterien. Es verbessert die Performanz auf der Zieldomäne signifikant und macht leistungsstarke Werkzeuge zugänglicher. Der Erfolg bei der Anpassung wirft die Frage auf, ob größere Foundation Models (FMs) inhärent überlegene, übertragbare Merkmale bieten. Folglich evaluiert der dritte Beitrag, PhenoBench, systematisch aktuelle Pathologie-FMs für komplexe Zellphänotypisierung. Mit dem granularen PhenoCell-Datensatz bewertet dieser Benchmark Merkmalsqualität und Generalisierungsfähigkeit von FMs über technische und medizinische Domänenverschiebungen hinweg. PhenoBench zeigt, dass FMs zwar Vorteile bei Datenknappheit bieten, aber keine Universallösung sind und spezifische Empfindlichkeiten aufweisen. Zusammen ergeben diese Beiträge eine kohärente Untersuchung, die von praktischer Anwendung und methodischer Verbesserung von Generalistenmodellen bis zur kritischen Bewertung von Foundation Models reicht und ein Toolkit für die dateneffiziente biomedizinische Bildanalyse bietet.Biomedical deep learning is often hindered by data scarcity and domain shifts. This monograph explores leveraging pre-trained models to address these challenges, presenting three interconnected contributions. First, the PEERCE pipeline demonstrates the practical value of this paradigm by combining fine-tuned networks with an off-the-shelf generalist model (Cellpose) for prediction of the PD-L1 Tumor Proportion Score in Angiosarcoma, a rare cancer. While successful, PEERCE highlights the need for careful pre-processing to adapt Cellpose and still requires some labeled data for fine-tuning other components. This limitation motivates the second contribution, SelfAdapt, a novel framework for unsupervised domain adaptation. SelfAdapt enables label-free adaptation of generalist models like Cellpose to new target domains using a student-teacher approach, L2-SP regularization, and label-free stopping criteria. It significantly improves model performance on the new domain, thereby making powerful tools more accessible. The success in adapting generalist models then naturally raises the question of whether even larger, more broadly pre-trained Foundation Models (FMs) offer inherently superior, transferable features. Consequently, the third contribution, PhenoBench, systematically evaluates contemporary pathology FMs for complex cell phenotyping. Introducing the granular PhenoCell dataset, this benchmark assesses FM feature quality and generalization capabilities across technical and medical domain shifts. PhenoBench reveals that while FMs offer advantages, particularly in low-data regimes, they do not yet constitute a universal solution and exhibit specific sensitivities. Together, these three contributions provide a cohesive investigation, progressing from practical application and methodological enhancement for leveraging generalist models to a critical evaluation of foundation models, offering a comprehensive toolkit for data-efficient biomedical image analysis
Dolomite and Mg Calcite as Mineral Thermometers in Mortar Binders. A High Resolution Raman Spectroscopic Study
No full textThis paper suggests the use of high‐resolution Raman scattering bands of MgCa carbonates as posteriori thermometer minerals in archaeometric studies. Therefore, the thermal behavior of two dolomite samples and the hydration and carbonation reaction in air of the decomposition products were investigated by Raman microspectroscopy. The increase in the calcination temperature resulted in the formation of – Raman silent MgO and – inert Mg calcite at 700°C–750°C. In contrast, the decarbonation, hydration, and recarbonation of sample material exposed to 750°C–900°C in a muffle furnace led to the appearance of Mg‐free calcite. High spectral resolution Raman spectroscopy enabled a spectral distinction between these two groups due to differences in the band parameters (peak position, bandwidth) of the vibrational (v1, v4, L) modes of calcite. In combination with Raman microspectroscopic mapping, this spectral information represents a new approach for the estimation of burning temperatures of medieval high‐fired gypsum mortars via natural dolomite impurities. Thus, the results of this work highlight the importance and potential of Raman microspectroscopy as a thermometric tool for elucidating the thermal history of anthropogenic fired materials, with potential applications for archaeometry and art technology, as well as for quality controls in the frame of the production of mineral mortar binders and ceramics or bricks, respectively.Deutsche Forschungsgemeinschaft (DFG) 10.13039/501100001659Interreg VI‐A Italy‐AustriaPeer Reviewe
Frühe Lesefähigkeiten und Vorläuferfähigkeiten des Schriftspracherwerbs deutschsprachiger und türkisch-deutschsprachiger Kinder: Disparitäten am Übergang vom Elementar- in den Primarbereich und Effekte der Förderung phonologischer Bewusstheit
No full textDie Dissertation untersucht frühe Leseteilfähigkeiten und schriftsprachliche Vorläuferfähigkeiten von türkisch-deutschsprachigen und monolingual deutschsprachigen Kindern am Übergang vom Elementar- in den Primarbereich. Teilstudie I analysiert, welche Leistungsunterschiede in den frühen Leseteilfähigkeiten und Vorläuferfähigkeiten zwischen den Sprachgruppen bestehen. Zudem wird geprüft, inwiefern die für monolinguale Kinder zentralen Vorläuferfähigkeiten des Leseerwerbs gleichermaßen bei türkisch-deutschsprachigen Kindern prädiktiv sind. Teilstudie II geht in einem quasi-experimentellen Prä-Post-Studiendesign der Frage nach, ob türkisch-deutschsprachige Kinder ebenso wie deutschsprachige Kinder (N = 420) von der Förderung der phonologischen Bewusstheit im Deutschen profitieren. Teilstudie I ergab, dass (schrift-)sprachliche Kompetenzunterschiede zwischen den Sprachgruppen bereits im Vorschulalter bestehen. Differenzielle Leistungsausprägungen zugunsten der deutschsprachigen Kinder zeigten sich vor dem Schuleintritt in Sprachkompetenzen und Vorläuferfähigkeiten, sowie am Ende des ersten Schuljahres in basalen Lesefähigkeiten und im Leseverständnis. Für die Vorhersage schulischer Lesefähigkeiten erwies sich die vorschulische Silben- und Reimbewusstheit in beiden Gruppen als zentraler Prädiktor. Teilstudie II belegt für beide Sprachgruppen positive Effekte einer phonologischen Förderung auf die Entwicklung von Vorläuferfähigkeiten. Hinsichtlich längerfristiger Effekte auf den Leseerwerb zeigten sich jedoch sprachgruppenspezifische Unterschiede, da lediglich die türkisch-deutschsprachigen Kinder in ihren basalen Lesefertigkeiten signifikant von der Förderung profitierten. Die Befunde der Dissertation sind für die pädagogische Praxis sehr relevant. Sie verweisen auf frühe Unterschiede im Leseerwerb und belegen das kompensatorische Potenzial phonologischer Förderung im Vorschulalter.The dissertation investigates early reading skills and early literacy skills of Turkish-German bilingual and monolingual German-speaking children at the transition from preschool to primary school. Substudy I addresses the differences in early literacy skills between the two language groups. In addition, it examines to what extent the skills that are central predictors of reading acquisition in monolingual children are equally predictive for Turkish-German bilinguals. Substudy II uses a quasi-experimental pre-post study design to analyze whether Turkish-German bilingual children benefit equally from a phonological training as German-speaking children (N = 420). Substudy I revealed that differences in (written) language competencies between the language groups already exist at preschool age. Differential performance in favor of German-speaking children was evident before school entry in language competencies and early literacy skills, as well as at the end of the first school year in basic reading skills and reading comprehension. Syllable and rhyme awareness assessed at preschool age proved to be a key predictor of later reading skills in both groups. Substudy II demonstrates positive effects of a phonological training on the development of early literacy skills for both language groups. However, with regard to longer-term effects on reading acquisition, group-specific differences emerged. Only Turkish-German bilingual children benefited significantly from the phonological training in terms of their basic reading skills. The findings of the dissertation are highly relevant for educational practice. They point to early differences in reading acquisition and highlight the compensatory potential of phonological training in preschool age
Von 'Ostdeutschen', 'Migrant:innen' und hegemonialem Deutschsein
No full textEmbargo bis 05.08.2026Auch im 35. Jahr der deutsch-deutschen Vereinigung stellt »der Osten« eines der spannungsgeladensten Phänomene der innerdeutschen Gegenwart dar. Entscheidend dafür ist dessen anhaltende Thematisierung im Rahmen des modern(isierungstheoretisch)en Paradigmas. Der Band plädiert deshalb für einen Paradigmenwechsel hin zu einer Kritischen Ostdeutschlandforschung, die ausgehend von einem anderen Blick auf die Welt, inklusive »des Westens«, einen anderen Blick auf »den Osten« sowie auf das Verhältnis von »Westen«, »Osten« und »dem (Post-)Migrantischen« ermöglicht. Etwaig Differentes muss dabei weder permanent problematisiert noch negiert werden, so dass Fragen von Identität und Alterität, von Einheit und Differenz angemessener als bisher adressiert werden können
Atomic-Scale Electric Potential Landscape across Molecularly Gated Bilayer MoS2 Resolved by Photoemission
No full textElectric gating in atomically thin field-effect devices based on transition-metal dichalcogenides has recently been employed to manipulate their excitonic states, even producing exotic phases of matter, such as an excitonic insulator or Bose–Einstein condensate. Here, we mimic the electric gating effect of a bilayer-MoS2 on graphite by charge transfer induced by the adsorption of molecular p- and n-type dopants. The electric fields produced are evaluated from the electronic energy-level realignment and Stark splitting determined by X-ray and UV photoelectron spectroscopy measurements and compare very well with literature values obtained by optical spectroscopy for similar systems. We then show that analysis of the inhomogeneous broadening and energy shifts of the quantum-well states of the valence band allows extraction of the full electric potential profile and charge-density redistribution across the entire heterojunction with atomic-scale precision, which is not accessible by other methods.Office of Naval Research 10.13039/100000006Alexander von Humboldt-Stiftung 10.13039/100005156Deutsche Forschungsgemeinschaft 10.13039/501100001659Peer Reviewe
Global thermal tolerance compilation for freshwater invertebrates and fish
No full textScientists have investigated the thermal tolerance of organisms for centuries, yet the field has not lost relevance as the environmental threats of thermal pollution and global change sharpen the need to understand the thermal vulnerability of organisms in landscapes increasingly subjected to multiple stressors. Freshwater fish and invertebrates are greatly underrepresented in recent large-scale compilations of thermal tolerance, despite the importance of freshwaters as a crucial resource and as havens for biodiversity. Therefore we compiled ThermoFresh, a thermal tolerance database for these organisms that includes literature from 1900 until the present, sourced from five languages to counteract geographic bias. The database contains over 6800 records for over 900 species, including 470 invertebrates, as well as 505 thermal tolerance tests conducted with additional stressors present. We provide a valuable resource to test hypotheses on thermal risks to freshwater organisms in present and future environments subject to multiple stressors.Deutsche Forschungsgemeinschaft (German Research Foundation)https://doi.org/10.13039/501100001659Chinese Academy of Sciences (CAS)https://doi.org/10.13039/501100002367Peer Reviewe
On a Hierarchy of Spectral Isomorphism Invariants
No full textWe consider a hierarchy of graph invariants that naturally extends the spectral invariants defined by Fürer (Lin. Alg. Appl. 2010) based on the angles formed by the set of standard basis vectors and their projections onto eigenspaces of the adjacency matrix. We provide a purely combinatorial characterization of this hierarchy in terms of the walk counts. This allows us to give a complete answer to Fürer's question about the strength of his invariants in distinguishing non-isomorphic graphs in comparison with the 2-dimensional Weisfeiler-Leman algorithm, extending the recent work of Rattan and Seppelt (SODA 2023). As another application of the characterization, we prove that almost all graphs are determined up to isomorphism in terms of the spectrum and the angles, which is of interest in view of the long-standing open problem whether almost all graphs are determined by their eigenvalues alone. Finally, we describe the exact relationship between the hierarchy and the Weisfeiler-Leman algorithms for small dimensions, as also some other important spectral characteristics of a graph such as the generalized and the main spectra.Peer Reviewe
Interactive room design as a tool for understanding form and style preferences
No full textGiven the substantial time people spend indoors, it is crucial to understand how aesthetic decisions in interior design are made, both to inform psychological research and guide practical applications. Recent studies show growing interest in form preferences, often suggesting a general preference for curved over angular shapes. Unlike previous research using predefined stimuli, our study allowed individuals to freely express their preferences in a living room through an interactive 3d web application. In an online experiment, participants ( N = 196) furnished a virtual living room according to their personal preferences, using a controlled set of furniture items that varied systematically in form (angular vs. curved) and style (modern vs. classic). Contrary to prior findings, we observed a small but significant overall preference for angular furniture. This effect was mainly driven by male participants, who selected more angular than curved items, while female participants showed no form preference. We found no general preference in terms of style. Additionally, we explored associations between design preferences and personality traits (BFI-2), but no significant relationships emerged. Overall, this study demonstrates the value of interactive 3d environments for investigating aesthetic preferences in a controlled yet flexible setting, allowing personalised design decisions beyond predefined conditions.Open Access funding enabled and organized by Projekt DEAL.Max Planck Institute for Human Development (2)Peer Reviewe
Audibilia
No full textThe contemporary debate about the ontology of sounds appears stuck. Two main contestants—the medial view and the located event view—each have a number of arguments on their side, while each needs to either accommodate or downplay the arguments speaking for the competing approach. In this paper, I want to argue that the two views are in fact not at odds with each other, but may, with some amendments, be combined in a two‐step model of sound perception inspired by Thomas Reid's theory of vision: Reid argues that we perceive bodies in virtue of perceiving visibles, two‐dimensional occupants of the visual field, best understood as relational properties. My suggestion is that we perceive sounds, understood as located events in virtue of perceiving audibles which have much in common with sounds as understood by medial theorists.Peer Reviewe
