259594 research outputs found
Sort by
The limits of one-size-fits-all patterns: a 5-step approach to testing homogeneity of within-person correlations
No full textExploring the role of teacher self-efficacy and personal environmental practices in integrating sustainability into teaching: a network analysis of German teachers
No full textIntegrating sustainability into school curricula is increasingly important, with teachers seen as key “change agents”. However, many lack specific preparation for Education for Sustainable Development (ESD), and there is considerable variability in how explicitly or implicitly they address these topics in their teaching. The purpose of this study was to investigate interpersonal and contextual factors related to ESD implementation, including self- and action-efficacy, personal attitudes, eco-anxiety, private engagement and knowledge, alongside perceived student interest and pressure, and school awareness. A total of 419 teachers from various German primary and secondary schools (M = 45 years, SD = 10.9; 68% female; teaching experience: M = 16 years, SD = 9.9) completed a cross-sectional online survey and knowledge test. Findings showed significant variation in how often teachers included sustainability in their teaching, unrelated to gender, school type, or training. Network analyses revealed that self-efficacy and private engagement—rather than teachers’ knowledge—were central predictors of ESD integration. Notably, private engagement emerged as a key bridge in the network, while high self-efficacy was closely tied to frequent classroom implementation. These results suggest that fostering teachers’ personal commitment and confidence may be more effective than focusing solely on knowledge to promote sustainability education
Secondary school students pursue differentiated achievement goals: the hexagonal achievement goal scale for students (HEX-S)
No full textOn the influence of social norms on individual achievement goals
No full textBackgroundIndividual achievement goals are influenced by the learning context, such as the classroom. In this social space, social norms emerge and shape motivation and behaviour. Classroom goal structures reflect injunctive norms (what is considered acceptable) and influence individual achievement goals. The role that descriptive norms (what others typically do or think) play in individual achievement goals is unclear. We propose that peer achievement goals reflect descriptive norms and additionally influence individual achievement goals.AimsWe aim to better understand contextual influences on individual student motivation by applying a social norms framework to study changes in individual achievement goals and acknowledge the role of peers.Sample and MethodsWe used longitudinal data from 4189 students from 169 classes at two time points after the transition to secondary school.ResultsWe calculated multilevel models to predict changes in individual mastery-, performance-approach, and performance-avoidance goals. As Level-2 predictors, class-level classroom goal structures represented injunctive norms, while peer achievement goals represented descriptive norms. Individual achievement goals and individual-level classroom goal structures were added on Level 1. Class-level classroom goal structures related to changes in individual achievement goals only if peer achievement goals were not added. If added on the classroom level, peer achievement goals remained as a single Level-2 predictor of changes in individual achievement goals.ConclusionWe demonstrated the key role that descriptive norms (reflected by peer achievement goals) play in individual achievement goals. The role of injunctive norms needs to be investigated further to enhance our understanding of how social norms shape individual student motivation
Fungal diversity and foliar disease dynamics in urban horse chestnut trees : a transect study across Southern to Northern European cities
No full textHorse chestnut trees (Aesculus hippocastanum) play a vital role in European urban landscapes, offering ecological, cultural, and aesthetic benefits. However, their health is increasingly affected by biotic stressors, including fungal pathogens and insect pests, within the context of urbanization and climate change. This study assessed leaf health in 659 horse chestnut trees across five European countries, using visual symptom scoring and fungal identification from necrotic leaf tissue. The results revealed a diverse fungal community associated with leaf blotch symptoms, indicating that multiple pathogens besides Guignardia aesculi are involved. Patterns of fungal symptoms and leaf miner (Cameraria ohridella) presence varied across geographic regions suggesting influence by broader environmental conditions. Fungal diversity was highest in both, the southern and the northern areas as compared to central European ones, with several identified fungal species having implications for leaf damage and public health. Street trees generally exhibited more severe foliar symptoms (higher Leaf Health Score) compared to park trees, an observation likely to more constrained growing conditions and increased exposure to urban stress. This study emphasizes the importance of integrating pathogen monitoring and species diversity into urban forest management to support the long-term resilience of urban tree populations
Light induced training of 3D printed mechanical metamaterials
No full textStimuli-responsive materials often react to changes in environmental conditions by altering their shape. Here, it is shown that even changes in materials that are not directly observable, such as local stiffening, can be exploited to introduce the concept of trainable materials. A fully 3D-printable filament based on thermoplastic polyurethane (TPU) functionalized with a bivalent crosslinker capable of undergoing a C,H insertion reaction under UV irradiation was developed. Specimens printed from this filament demonstrate a gradual increase in stiffness, reaching almost 300% of their initial stiffness after 50 hours of irradiation. To exploit this tunability, mechanical metamaterials incorporating the developed material are engineered. By utilizing an instability-driven transformation under compression, it is demonstrated how local stiffening can be amplified through rational design. Moreover, by exploiting the unusual mechanical behavior of metamaterials arising from their internal architecture, a closed-loop system is presented in which, under compressive load, the metamaterial closes an electric circuit that activates UV light, which in turn modifies the properties of the base material. Through this approach, two trainable systems are realized: one that progressively conforms its shape to mechanical compression, and another that gradually increases its resistance to an applied force, mimicking the physical training of biological tissues
Investigating Choices Regarding the Accuracy-Transparency Trade-Off of AI-Based Systems Across Contexts
No full textTowards robust, efficient and reproducible neural architecture search
No full textThe advent of deep learning has revolutionized artificial intelligence (AI), enabling breakthroughs in a plethora of real-world applications. However, manually designing high-performing neural network architectures remains a complex and expensive task, often requiring expert knowledge and extensive trial-and-error evaluations. Neural Architecture Search (NAS) has emerged as a promising solution to automate this manual process, leveraging search algorithms to discover new architectures that outperform the human-designed counterparts.Despite its potential, NAS faces significant challenges that hinder its widespread adoption. First, training large architectures over vast search spaces is prohibitively expensive, limiting its accessibility in research and industry, and at the same time making reproducibility more difficult. Addressing this challenge is key to democratizing NAS and enabling deployment in resource-constrained environments. Furthermore, the lack of standardized benchmarks and reproducibility challenges hinder fair comparisons, eventually slowing progress. Second, many NAS methods prioritize search efficiency at the expense of robustness, leading to suboptimal architectures that perform poorly across different datasets or conditions. A reliable NAS framework must balance efficiency and robustness to ensure that the found architectures perform well on unseen data. Third, NAS struggles to optimize architectures for multiple objectives, for instance, calibration during distribution shift or hardware efficiency, making real-world deployment difficult. Models must be reliable, energy-efficient, and meet latency constraints under varying conditions to be practical for deployment.This dissertation addresses these challenges by introducing novel methods, frameworks, and benchmarks that enhance computational efficiency, robustness, and reproducibility of NAS algorithms, ultimately driving NAS toward more realistic and reliable solutions. It first formalizes the NAS problem and provides a comprehensive survey of existing work. Next, it advances reproducibility through tabular and surrogate NAS benchmarks, benchmark suites, and open-source tools. Then, it further develops methods to improve the robustness of NAS algorithms, both in-distribution and under distribution shifts. Finally, it introduces multi-objective gradient-based optimization techniques to create architectures that are both high-performing and computationally efficient.By tackling these fundamental challenges, this dissertation contributes to making NAS more accessible, robust, and practical, enabling the discovery of architectures that balance various objectives across diverse domains. Its findings have broad implications for AutoML and deep learning, fostering the development of scalable AI models while driving innovation in the field
