2972 research outputs found
Sort by
Efficient and Controllable Model Compression through Sequential Knowledge Distillation and Pruning
No full textEfficient model deployment is a key focus in deep learning. This has led to the exploration of methods such as knowledge distillation and network pruning to compress models and increase their performance. In this study, we investigate the potential synergy between knowledge distillation and network pruning to achieve optimal model efficiency and improved generalization. We introduce an innovative framework for model compression that combines knowledge distillation, pruning, and fine-tuning to achieve enhanced compression while providing control over the degree of compactness. Our research is conducted on popular datasets, CIFAR-10 and CIFAR-100, employing diverse model architectures, including ResNet, DenseNet, and EfficientNet. We could calibrate the amount of compression achieved. This allows us to produce models with different degrees of compression while still being just as accurate, or even better. Notably, we demonstrate its efficacy by producing two compressed variants of ResNet 101: ResNet 50 and ResNet 18. Our results reveal intriguing findings. In most cases, the pruned and distilled student models exhibit comparable or superior accuracy to the distilled student models while utilizing significantly fewer parameters
Illness Perceptions of Patients with Occupational Skin Diseases in a Healthcare Centre for Tertiary Prevention: A Cross-Sectional Study
No full textObjectives: To investigate the illness perceptions of patients with occupational skin diseases (OSDs). Design: Cross-sectional study. Setting: Specialised healthcare centre for inpatient and outpatient individual prevention in occupational dermatology in Germany. Participants: A total of 248 patients with hand eczema (55.2% female; average age: 48.5 years, SD: 11.9) were included in the final analyses. Measures: A modified and recently validated version of the ‘Revised Illness Perception Questionnaire’ (IPQ-R) was used to assess illness perceptions. Severity of skin disease was evaluated with the Patient-Oriented Eczema Measure (POEM), the Osnabrueck Hand Eczema Severity Index (OHSI), and a single, self-reported global item. The Erlangen Atopy Score (EAS) was used for atopy screening. Results: We found strong illness identity, high emotional impact, and long timeline beliefs, meaning that study participants perceive their OSD on the hands as a highly symptomatic, emotionally burdening, and chronic condition. Results suggest that hand eczema has a major impact on how participants manage their own lives, particularly during everyday life and occupational activities. Study participants predominantly identified irritant or sensitising substances and activities at work as well as skin protection regimes as causes of their disease. Conclusions: Healthcare workers should consider the illness perceptions as well as the disease burden of patients with an OSD on the hands in clinical practice. Multi-professional approaches to patient care should be sought. Illness perception in (occupational) dermatological patients should be the subject of further research
Eyes on the road: brain computer interfaces and cognitive distraction in traff
Get PDFNovel wearable neurotechnology is able to provide insight into its wearer's cognitive processes and offers ways to change or enhance their capacities. Moreover, it offers the promise of hands-free device control. These brain-computer interfaces are likely to become an everyday technology in the near future, due to their increasing accessibility and affordability. We, therefore, must anticipate their impact, not only on society and individuals broadly but also more specifically on sectors such as traffic and transport. In an economy where attention is increasingly becoming a scarce good, these innovations may present both opportunities and challenges for daily activities that require focus, such as driving and cycling. Here, we argue that their development carries a dual risk. Firstly, BCI-based devices may match or further increase the intensity of cognitive human-technology interaction over the current hands-free communication devices which, despite being widely accepted, are well-known for introducing a significant amount of cognitive load and distraction. Secondly, BCI-based devices will be typically harder than hands-free devices to both visually detect (e.g., how can law enforcement check when these extremely small and well-integrated devices are used?) and restrain in their use (e.g., how do we prevent users from using such neurotechnologies without breaching personal integrity and privacy?). Their use in traffic should be anticipated by researchers, engineers, and policymakers, in order to ensure the safety of all road users
Detecting Historical Terrain Anomalies With UAV-LiDAR Data Using Spline-Approximation and Support Vector Machines
No full textThe documentation of historical remains and cultural heritage is of great importance to preserve historical knowledge. Many studies use low-resolution airplane-based laser scanning and manual interpretation for this purpose. In this study, a concept to automatically detect terrain anomalies in a historical conflict landscape using high-resolution UAV-LiDAR data was developed. We applied different ground filter algorithms and included a spline-based approximation step in order to improve the removal of low vegetation. Due to the absence of comprehensive labeled training data, a one-class support vector machine algorithm was used in an unsupervised manner in order to automatically detect the terrain anomalies. We applied our approach in a study site with different densities of low vegetation. The morphological ground filter was the most suitable when dense near-ground vegetation is present. However, with the use of the spline-based processing step, all filters used could be significantly improved in terms of the F1-score of the classification results. It increased by up to 42% points in the area with dense low vegetation and by up to 14% points in the area with sparse low vegetation. The completeness (recall) reached maximum values of 0.8 and 1.0, respectively, when taking into account the results leading to the highest F1-score for each filter. Therefore, our concept can support on-site field prospection
Theoretische Untersuchung zum Misch-Netzwerkbildner-Effekt zur Optimierung von Glaselektrolyten
No full textBei Mischung von Netzwerkbildnern in ionenleitenden Gläsern wird die Leitfähigkeit beweglicher Ionen modifiziert und kann insbesondere gesteigert werden. Die beweglichen Ionen in Gläsern sind zumeist Alkaliionen, in der praktischen Anwendung insbesondere Lithium- oder Natriumionen. Bei Erhöhung ionischer Leitfähigkeiten lässt sich der Misch-Netzwerkbildner-Effekt zur Optimierung von Batterien und zur Verbesserung von Glaselektrolyten ausnutzen, die in chemischen Sensoren, intelligenten Fensterscheiben und Superkondensatoren Verwendung finden. Zur Materialoptimierung ist von entscheidender Bedeutung, ein besseres theoretisches Verständnis des Effektes zu gewinnen. In dieser Arbeit wird davon ausgegangen, dass die Verteilung der Anionenladung (Gegenladung) durch die Bildung unterschiedlich chemischer Netzwerkbildner-Einheiten in Gläsern geändert wird. Diese Netzwerkbildner-Einheiten lassen sich experimentell aus Magic-Angle-Spinning-, NMR- und Raman-Spektren bestimmen. Für die theoretische Modellierung der gemessenen Konzentrationen der Netzwerkbildner-Einheiten werden thermodynamische Modelle entwickelt. Aufbauend auf dieser Modellierung wird der Ionentransport untersucht. Um die Modifikation der Energielandschaft, die mit der Änderung der Konzentrationen der Netzwerkbildner-Einheiten verbunden ist, für den langreichweitigen Ionentransport zu beschreiben, wird angenommen, dass nicht-brückende Sauerstoffatome lokalisierte Coulomb-Fallen für die beweglichen Ionen bilden. Auf diese Weise wird die Energielandschaft für den langreichweitigen Ionentransport durch die Änderung der Konzentrationen der Netzwerkbildner-Einheiten modifiziert. Unter Zugrundelegung dieses Konzepts werden mit Monte-Carlo-Simulationen des thermisch aktivierten Transports der mobilen Ionen sowie perkolationstheoretischen Methoden ionische Leitfähigkeiten und ihre Aktivierungsenergien bestimmt. Vergleiche mit Experimenten für verschiedene ionenleitende Gläser liefern eine gute Übereinstimmung mit den gemessenen Änderungen in Abhängigkeit von Mischungsverhältnis von Netzwerkbildnern
The Social Dynamics of Collaboration in Environmental Governance and Management
No full textDoing justice to local knowledge and contexts in sustainability transformations requires multi-actor collaboration and broad stakeholder participation in environmental governance and management. Assumptions from network and collaborative governance have therefore led to the rise of co-management, in which decision-making power and management responsibility are shared among state and non-state actors interacting in knowledge partnerships. Yet despite normative claims, key challenges for many co-management approaches – especially in the environmental realm – remain to (1) navigate through tensions over meaning and competing narratives, (2) deal with blurred roles of authority and responsibilities due to decentralization, and (3) manage socio-historical pasts, where cooperation and conflict are entangled in actor relationships. Over the years, scientific communities debating the suitability of these collaborative processes have warned about viewing collaborative approaches as a magic bullet for targeting environmental problems. This thesis seeks to contribute to the debate and adds a complementary perspective on collaborative governance and co-management arrangements by examining the context-dependent social dynamics that arise when a group of actors involved in collaborative approaches negotiate and implement environmental governance and management measures. It therefore asks: How do the social dynamics between actors shape governance networks and influence collaborative governance arrangements?
The first part of this thesis addresses the social dynamics of movements toward sustainable futures and related narratives of vision and identity. It argues that narratives regarding vision and identity accompany sustainability transitions and collective behavior change and that these influence and reflect social dynamics on a broader scale. Yet narratives are often reduced to a shorthand – an abbreviated (though affective) narrative expression such as a slogan, song, dance or image, which is memorable and readily communicable across the community and beyond. These abbreviated forms are referred to as concise affective narrative expressions (CANEs), which consist of a characteristic piece extracted from the complete narrative for a memorable, easily communicable, and affective verbal or visual representation of a core message. Furthermore, the challenges of collaboration described above became evident in the case study. The case study describes a regional cooperation consisting of different state actors and stakeholders from agriculture, forestry, water management and hunting. These actors discuss how to implement the Natura 2000 regulation in current land and forest management. The social dynamics of collaboration in the case were investigated by means of an interdisciplinary conceptual framework based on narrative and social network theory called the relational narrative approach. This framework is built on the assumption that the social relational structure between actors, and the stories they tell, is a co-production of narratives and dynamics at the group level. The mechanisms that influence emerging dynamics are (1) the interplay between collaborative relationships and narrative congruence between individual actors, (2) the characteristics of actors, and (3) the actors’ embeddedness in the wider social structure (which will be detailed in Paper II, presented as part of this thesis).
The idea of narrative congruence is detailed in Paper III with the aim of exploring the phenomenon of a common narrative and to examine which social drivers shape the emergence of a common narrative among diverse actors involved in co-management. The argument in this part suggests that frequent interaction between two actors and a trusted leader with many reciprocal ties of trust are significant drivers that support the emergence of narrative congruence.
Despite regular interaction between participants in the regional cooperation, the findings set out in Paper IV imply that these actors are unable to co-create a common narrative that would break the patterns of conflict and antagonistic perceptions of identity. Instead of a common narrative that would presumably facilitate the development of collaborative ties between agents, two opposing narratives are reproduced that vie with each other over power and competencies when it comes to appropriate management planning in the Natura 2000 areas – thus generating an “us versus them” dynamic. This polarization into two opposing sub-groups is however not transferred to the relationships that participants of this particular regional cooperation initiative have with one another. On the contrary, the regional cooperation is a network “supported by many shoulders”, the actors involved are familiar with each other and several coordinators and mediators among the actors ensure that a great deal of exchange occurs among participants.
This thesis concludes by discussing three insights gleaned from this research undertaking. First, social dynamics are ubiquitous and intangible phenomena whose mechanisms influence multi-actor interaction in collaborative governance and management. Second, these intangible forces can be studied, explained and made manifest by way of the narratives that actors tell and the social relations and societal embeddedness of actors. Third, common narratives evolve around a trusted leader with many ties of trust to other actors and require frequent encounters and long-term nurturing in order to arise from multi-actor collaborations
Magnetogenetic Control of Cellular Functions using Biofunctionalized Magnetic Nanoparticles Inside Living Cells
Get PDFRemote control of cellular functions using magnetic forces offers unique opportunities in fundamental research and biomedical applications. The intracellular application of functionalized magnetic nanoparticles (MNP) provides the possibility to spatiotemporally increase the concentration of specific target proteins and thus locally enhance protein interactions (space mode magnetogenetic control). These can be exploited for site-specific activation of signaling pathways, for instance by increasing reaction turnovers or by inducing phase separation. However, designing MNPs suitable for effective, unbiased manipulation of target proteins inside cells has remained challenging.
This work aimed to design biofunctionalization of MNPs for space mode magnetic manipulation of downstream signaling effector proteins at the plasma membrane and in the cytosol. For initial proof-of-concept experiments, application of previously established magnetic intracellular stealth MNPs based on natural ferritin (MagIcS) as a highly biocompatible protein cage were explored. As these exhibited limitations in terms of magnetic properties, a new MNP design was implemented based on a single-step surface coating of synthetic magnetic core nanoparticles with green fluorescent protein (GFP) fused to the iron binding site of Mms6 from magnetotactic bacteria (syMagIcS). In doing so, a stable biocompatible coating was created, which simultaneously enables site-specific recruitment of proteins of interest to the magnetic nanoparticle surface. The direct functionalization of synthetic magnetic cores yielded optimized magnetic properties and offers the possibility to customize core sizes and thus further enhance magnetic responsiveness. Exploiting the coherent functional design of MagIcS and syMagIcS, applicability for intracellular magnetogenetic use in space mode was explored at different levels. Efficient in situ MNP biofunctionalization with intracellular effector proteins by direct capturing via GFP nanobodies was achieved and intracellular translocation by using magnetic field gradients was obtained for both types of MNP. Activation of G-proteins at the plasma membrane was achieved by magnetogenetic translocation of MNP-bound catalytically active region of GEF proteins, with syMagIcS clearly showing superior performance as compared to MagIcS. Furthermore, liquid-liquid phase separation of the intrinsically disordered protein Dvl2, which plays an important role in Wnt signaling pathways, was induced by space mode magnetogenetics. With this tool at hand, we aim to understand the activation and regulatory mechanisms of these signaling pathways in more detail and test hypotheses by manipulating the signaling activation. These proof-of-concept experiments highlight the exciting possibilities of space mode magnetic manipulation to explore the spatiotemporal regulation of cellular processes in living cells and open new avenues in regenerative medicine
Exploring Neural Dynamics of Performance Monitoring: A Comprehensive Series of EEG Studies on Cognitive Control across Contexts
Get PDFEvery human behavior bears the potential for a mistake. Sometimes it is hard to make the right decision. Sometimes it is difficult to perform an action correctly. Our everyday life involves a constant risk of performing a behavior more or less incorrectly, leading to negative consequences of varying severity. To err is human, but so is learning from our actions. In the course of evolution, humans have developed a complex neural system to detect errors, resolve cognitive conflict and adapt future behavior. As multifaceted the range of situations and behaviors we encounter in everyday life is, so are the neuroscientific studies and findings on this performance monitoring system.
Therefore, this dissertation aims to integrate different approaches to investigate performance monitoring with a particular focus on a neural indicator that consistently appears in the literature on performance monitoring research, namely frontomedial theta activity (FMT). In a series of four studies, FMT activity in the scalp-recorded electroencephalogram (EEG) was examined in different situations and settings, applying different analysis procedures and considering interpersonal differences and intraindividual temporal stability. Study 1 investigated fine-grained additive effects of FMT activity in response to multiple independent outcomes in a non-dynamic laboratory setting and their stability over time. In contrast to the distinctive feedback presentation in study 1, study 2 investigated the effects of continuously incoming information during the performance monitoring process in a dynamic shooting task within virtual reality. As in a wide variety of daily actions, this design allowed for the anticipation of an outcome before it occurs and thus “online” performance monitoring. Study 3 advances the dynamics of the setting even further by implementing a shooting task without virtual reality but in a Mobile Brain/Body Imaging (MoBI) setting (Makeig et al., 2009), using toy guns with foam darts to shoot a stationary target. This study investigated the characteristics of the oscillatory phase of the measured FMT signal. Study 4 investigated FMT activity in response to cognitive conflict rather than negative action outcomes. Using source separation analyses, individual sensitivities to different kinds of conflict are investigated to determine whether FMT has the same or different neural sources across different contexts.
The four studies are discussed to determine the functional role of FMT in performance monitoring and whether FMT may reflect a unitary signal for performance monitoring across diverse contexts and tasks
Flipped Classroom and Learning Analytics in Higher Education: Effective Development, Integration, and Learner Support
No full textThis dissertation investigates the transformation of higher education in response to evolving workforce demands. As universities globally transition to diverse teaching formats, the integration of Flipped Classroom (FC) courses and the utilization of Learning Analytics (LA) take center stage. The FC approach, emphasizing interactive in-class sessions and pre-class self-learning phases, offers opportunities for skill development aligned with changing workforce needs. Challenges in FC design, however, persist. Learning Analytics emerges as a solution to enhance self-regulated learning skills through data collection and analysis. The dissertation aims to systematically develop and integrate FCs into higher education operations, proposing a process model, and explores the potential of LA to optimize FC courses. The research contributes insights, implications, and avenues for further exploration in the dynamic landscape of contemporary education
Report about the symposium on “Refugees and Peace: Situated Knowledge, Lived Experience, and Gender Dynamics”
Get PDFThe symposium placed the focus on peace and displacement. Whereas scholars in Forced Migration Studies have extensively examined the relation between conflict and displacement, peace has widely been neglected thus far. The symposium and its associated research project sought to address this gap by exploring the various meanings that displaced people ascribe to peace and ways in which they contribute to peaceful conditions. The following report summarizes key points and outcomes that were discussed during the symposium