Technische Universität Dresden: Qucosa
Not a member yet
16344 research outputs found
Sort by
Optimization of Sputtered Hafnium Zirconium Oxide Ferroelectrics for Memory Applications
No full textSince the initial publication of ferroelectricity in HfO2 thin films in 2011, numerous theories and models have been proposed to elucidate the root of this phenomenon. It is widely accepted that the ferroelectricity in HfO2 arises from its polar orthorhombic crystal phase, which is metastable at room temperature and under standard atmospheric conditions. Therefore, sputtering, being a non-equilibrium process, can be advantageous in stabilizing the metastable crystal structure of HfO2 in thin film. With regard to the practical processing, the incorporation of ferroelectric (FE) HfO2 in the front-end-of-line (FEOL) requires complicated steps and significant cost. Consequently, the back-end-of-line (BEOL) integration of FE HfO2 is technically preferred. However, this straightforward integration is accompanied by a trade-off, which is the limited thermal budget to prevent the detrimental impact on the components finished in FEOL processes. According to existing research, sputtered HfO2 requires high temperature (> 600 °C) anneals to activate a proper FE switching. Thus, the primary objective of this study is to engineer the sputtered hafnium zirconium oxide (HZO) to be thermally compatible with BEOL processes (< 450 °C). Through the optimization process, the FE response of HZO is found to be highly dependent on the sputtering parameters. This indicates that distinct optimal sputtering processes must be employed in accordance with varying sputter target conditions. By summarizing the effects of the sputtering parameters, general models are proposed to guide the optimization of sputtered HZO. Moreover, the reproducibility of similar ferroelectricity is demonstrated on the sputter targets with different conditions and from various suppliers. Further enhancements in the ferroelectricity of the ferroelectric capacitor are achieved via electrode engineering. Based on the results of this work, sputtered HZO promises an expedient and straightforward BEOL integration of FEs.:Contents
List of Figures vi
List of Tables xiv
List of Abbreviations xv
List of Symbols xvii
1 Introduction 1
2 Theoretical Background 3
2.1 Ferroelectricity and Ferroelectric Memory Device 3
2.1.1 Basics of Ferroelectricity 3
2.1.2 Ferroelectric Field-Effect Transistor 5
2.1.3 Ferroelectric Random-Access Memory 6
2.2 Hafnium-oxide-based Ferroelectrics 8
2.2.1 Hafnium Oxide 8
2.2.2 Doped Hafnium Oxide 10
2.2.3 Zirconium-doped Hafnium Oxide 11
2.2.4 Role of the Electrode 12
2.3 Fabrication Technology for HfO2-based Ferroelectrics 13
2.3.1 Atomic Layer Deposition 13
2.3.2 Sputtering 15
3 Experimental Section 18
3.1 Fabrication 18
3.1.1 Capacitor Stack Deposition 18
3.1.2 Capacitor Annealing 20
3.1.3 Capacitor Patterning 20
3.2 Characterization 22
3.2.1 Structural and Chemical Analysis 22
3.2.2 Electrical Analysis 27
4 Path of Sputtered Hafnium-Zirconium Oxide to Back-End-of-Line Compatibility 33
4.1 Effect of Sputtering Parameters 33
4.1.1 Sputtering Power 33
4.1.2 Zirconium Oxide Content 39
4.1.3 Sputtering Pressure 45
4.1.4 Other Parameters 54
4.2 Towards Back-End-of-Line Compatibility 60
4.2.1 High Remanent Polarization after BEOL-compatible Annealing 60
4.2.2 General Model 62
5 Optimizing the Sputtered Hafnium-Zirconium Oxide Ferroelectric Capacitor 65
5.1 Titanium Nitride Electrode Engineering 65
5.1.1 Effect of Nitrogen Tuning on Titanium Nitride 65
5.1.2 Engineering Oxygen Scavenging 68
5.2 Tungsten Oxide Electrode Engineering 74
5.2.1 Effect of Oxygen Tuning on Tungsten Oxide 75
5.2.2 Engineering Oxygen Supply 79
6 Conclusions and Outlook 85
7 Bibliography 8
Forestry Education Reimagined: Reform and Best Practices for the Future
No full textForestry education stands at a crossroads. The classical model – rooted in centuries-old tradi-tions – faces growing scrutiny amidst rapidly evolving societal, environmental, and economic demands. Today, technological advancements, especially in artificial intelligence (AI), add an-other layer of opportunity and challenge. Is classical forestry education reformable to meet these modern demands? Can we harness AI alongside proven best practices to ensure its rele-vance in a data-driven world?
Forests play a critical role in addressing global challenges, including climate change, biodiver-sity conservation, sustainable development, and the bioeconomy. At the same time, the rise of AI and digital tools is reshaping educational paradigms and offering innovative ways to analyze, simulate, and manage natural resources. Forestry education must adapt – not only to equip future foresters with traditional skills but also to empower them with competencies in AI, data analytics, and machine learning. This integration is essential for navigating complex environ-mental challenges and restoring trust and relevance within society.
This conference invites students, educators, and practitioners to explore and share perspec-tives on:
Reforming Traditional Models: How can established forestry education frameworks be up-dated to incorporate AI and other technological innovations?
Exemplary Practices: What are the best practices in curricula, teaching methods, and stake-holder collaboration that effectively blend traditional knowledge with AI-driven insights?
Societal Alignment: How can forestry education align with contemporary societal expectations and policy demands using AI-powered analytics and predictive modelling?
Interdisciplinary Competencies: What new skills—including AI, data science, and environmen-tal informatics—are essential for addressing modern forestry challenges?
Reclaiming Social License: In what ways can education, bolstered by transparent, data-driven AI tools, help reclaim forestry’s social license to operate?
By integrating AI into the conversation, this conference aims to chart a course for a future where forestry education not only preserves its rich heritage but also leverages cutting-edge technology to tackle global challenges more effectively.:1 Foreword 1
2 Ethical aspects of the use of AI in academia 4
3 Artificial Intelligence as a Tool for Strengthening Social License in Social Forestry: Towards Transparent and Participatory Forest Governance 5
4 Application of Artificial Intelligence in Education and Research at the Faculty of Forestry, University of Sopron 6
5 Vertical Harmonization of Forestry Education in Hungary 7
6 Green Business dual degree programme 8
7 Mediterranean Forestry and Natural Resources Management (MEDfOR) 9
8 The use of AI in Study and Science 10
9 Open Educational Resources (OER) and AI – the essence of quality 11
10 Proposals for European Higher Education Cooperation within the SILVA Network 12
11 The Use of Artificial Intelligence in Academic Education Experiences at the University of Göttingen, Germany 13
12 Discussion Results 14
13 Group photos 15
14 Change of presidency 1
Instruction Versus Experience in Rapid Learning: A Multimodal Approach to Understanding Cognitive and Neural Adaptations
No full textThis dissertation investigates the neural mechanisms and cognitive processes that support learning in its earliest stages under instruction and exploration. While real-world learning often interweaves these modes, this work disentangles their contributions by isolating the unique features of each, introducing a third condition of interest. Instruction-based learning speed and efficiency provide a unique lens to study the timescale of learning, especially compared to the slower, feedback-driven processes of trial-and-error and passive observation. By accounting for these temporal differences, this work addresses a key gap in the literature: understanding how the brain achieves the efficiency of instruction-based learning compared to trial-and-error learning. The three core studies use complementary exploratory and hypothesis-driven methodologies, and together, they contribute to a broader understanding of learning as a general cognitive domain.
Study 1 compared behavior, brain activity, and representational dynamics across instruction-based, observation-based, and trial-and-error learning. Behaviorally, instruction-based learning outperformed feedback-driven approaches, whether via active exploration or passive observation. Active exploration proved more effective than passive observation, which, in turn, led to the poorest rule implementation. This finding suggests that active exploration enhances feedback utilization. Univariate analyses revealed distinct cognitive demands across learning modes. In trial-and-error learning, cognitive control regions were more engaged early in learning and showed a steeper decrease in activity as rule exploration progressed, reflecting progressive task automation. In contrast, instruction-based learning was characterized by a faster increase in activity within the default-mode network (DMN), suggesting the rapid internalization and consolidation of task rules. Additionally, ventral striatal activity increased rapidly during instructed learning trials, confirming the specificity of this pattern to instruction-based learning. This extends previous findings, which had linked ventral striatal activity to implementation trials following a one-time instruction, by demonstrating its role during the learning stage. Ventral striatal activity was interpreted as reflecting the internal motivation associated with covertly implementing task rules as instructed, highlighting the rapid engagement of reward-related processes to reinforce newly acquired associations. This interpretation was reinforced by multivariate analyses, which revealed response preparation in sensormotor regions during instructed trials before their implementation. The multivariate approach was specifically designed to compare the representation of individual stimulus-response (S-R) rules across the learning and implementation stages of the task. Results showed that individual rule representations were present in prefrontal and parietal regions across conditions and remained stable across stages. This stability suggests that action-ready representations are rapidly established across learning modes, probably facilitating effective implementation. Notably, this finding was held even when implementation was not required during learning, as in the instruction- and observation-based learning conditions. In contrast, in sensorimotor regions, response representations were evident during trial-and-error learning trials, where responses were overtly implemented, and during instructed learning trials, where they reflected covert response implementation. Passive observation of correct and incorrect task rules did not generate similar sensor motor representations during learning. This suggests that passive learning via feedback hindered covert practice and may have contributed to the poorest performance during implementation. The combination of covert practice in instruction-based learning and this learning mode's overall lower cognitive demand was interpreted as a key factor underlying its behavioral advantage.
Study 2 identified distinct connectivity patterns for each learning mode. Reward-based and salience mechanisms were evident in feedback-based learning when driven by active exploration but not during passive observation. Connectivity patterns involving regions associated with inhibition suggested that inhibitory mechanisms may have suppressed reward-related processes in the observation condition. Notably, stronger inhibitory mechanisms during passive observation were associated with better performance in implementation trials, suggesting that inhibitory control may help optimize task performance when feedback is passively received. In trial-and-error learning, connectivity patterns involving the salience, default-mode, and attentional network were interpreted as reflecting the role of the salience network in reallocating attention to critical events when these are detected, supporting the flexibility required for this learning mode. Study 2 also involved the cerebellum, notably lobule VI, linked to the salience network and its function in switching between the default-mode and executive control networks. This finding is significant as it positions this work as one of the first to highlight cerebellar involvement in reward-based learning in humans. Finally, Study 2 identified frontostriatal mechanisms associated with instruction-based learning, although the observed connectivity patterns diverged from previous reports. These discrepancies were attributed to differences in experimental design and the varying cognitive control demands across rapid learning paradigms. The involvement of striatal regions across Study 1 and Study 2, specifically during instruction-based learning, confirmed the uniqueness of these activity patterns to this learning mode. This finding is consistent with prior research on rapid learning paradigms. In particular, frontostriatal mechanisms reflect the rapid neural changes characteristic of instruction-based learning, suggesting that the striatum plays a critical role in the rapid acquisition and implementation of new information. This striatal involvement may underpin the distinct advantage of instruction-based learning, where individuals can quickly internalize rules and efficiently execute responses.
Study 3 used an online TMS protocol to examine the causal roles of default-mode and cognitive control regions (the right angular gyrus, rAG, and the right middle frontal gyrus, rMFG) in instruction-based and trial-and-error learning. Based on univariate findings from Study 1, it was hypothesized that perturbation of either region at different time windows would differentially disrupt learning conditions. However, significant effects were observed only in the rAG group. The lack of significant group differences limits definitive conclusions about the regional specificity and direction of these effects, suggesting that neither region is uniquely or causally involved in either learning mode. Despite deviations from the pre-registered sample size and experimental groups, this study represents a first step in using online TMS to compare learning via instruction and exploration. Consistent with Study 1, Study 3 also confirmed the behavioral advantage of instruction-based over feedback-based learning. This finding reinforces that explicit instructions facilitate learning by reducing cognitive load, providing clear guidance, and bypassing the need for exploratory processes required in feedback-based approaches.
Finally, Studies 1, 2, and potentially study 3 highlighted common mechanisms that underpin learning across modes. These mechanisms appear to reflect general brain processes supporting task performance. Specifically, they indicate a shift towards more automatic or consolidated processing as participants become familiar with the task rather than mechanisms unique to the learning process. These results are crucial, as they demonstrate broader network dynamics that facilitate task performance. Study 1 observed increased DMN activity and decreased frontoparietal (FP) activity, while Study 2 showed increased within-DMN connectivity, decreased within-FPN connectivity, and decreased DMN-FPN connectivity. In Study 3, no performance differences were observed within each learning mode after disruption of DMN- or FPN-related regions, suggesting that neither network plays a uniquely causal role in the learning process, at least under the specific conditions tested. These findings demonstrate how the brain can share mechanisms to support general task performance while flexibly adapting to distinct learning conditions with different contextual and temporal characteristics, as reflected in condition-specific brain activity patterns.:1. SUMMARY 1
2. PREFACE 6
3. INTRODUCTION 8
3.1. THEORETICAL FRAMEWORKS OF LEARNING: BEHAVIORAL, COMPUTATIONAL AND PSYCHOLOGICAL THEORIES 8
3.2. NEUROBIOLOGY OF LEARNING 14
3.2.1. Dopaminergic pathways 14
3.3. NEUROIMAGING STUDIES 18
3.3.1. Instructed Reinforcement Learning paradigms: Merging exploration with instruction 19
3.3.2. Learning rules via instruction versus exploration: The importance of the timescale 22
3.3.3. Tracking Rapid Learning trial by trial: Paradigm and studies 24
3.4. PREMISES TO STUDY 1: BRAIN REPRESENTATIONS DURING INITIAL LEARNING AND SUBSEQUENT TASK EFFICIENCY 27
3.4.1. A methodological note: MVPA bias in learning paradigms 30
3.5. PREMISES TO STUDY 2: NETWORK DYNAMICS DURING INITIAL LEARNING AND SUBSEQUENT TASK EFFICIENCY 32
3.6. PREMISES TO STUDY 3: TESTING THE CAUSAL INVOLVEMENT OF CORTICAL REGIONS DURING INITIAL LEARNING VIA INSTRUCTION AND EXPLORATION 33
3.7. INITIAL LEARNING: THE EXPERIMENTAL PARADIGM 33
3.7.1. Behavioral pilot study 35
4. STUDY 1: INITIAL LEARNING IN THE BRAIN: FROM RULES TO ACTIONS 39
4.1. SUMMARY 39
4.2. INTRODUCTION 39
4.3. METHOD 43
4.3.1. Participants 43
4.3.2. FMRI paradigm 44
4.3.3. Data acquisition 48
4.3.4. Data processing 48
4.3.5. Data analysis 50
4.4. RESULTS 55
4.4.1. Behavior 55
4.4.2. Univariate fMRI analysis 57
4.4.3. Learning condition-specific signal change 63
4.4.4. MVPA 67
4.5. DISCUSSION 79
4.5.1. The advantage of explicit instructions on learning: Early proceduralization of instructed S-R rules 80
4.5.2. Mean neural change during learning reflects decreasing cognitive control demand 82
4.5.3. Representational dynamics in prefrontal, parietal and premotor cortex 86
4.6. CONCLUSIONS 88
5. STUDY 2: THE CONNECTED LEARNING BRAIN 89
5.1. NOTE TO STUDY 2 89
5.2. INTRODUCTION 90
5.3. METHOD 91
5.3.1. Data processing 92
5.3.2. Data analysis 94
5.4. RESULTS 97
5.4.1. Linear connectivity change during learning 97
5.5. DISCUSSION 109
5.5.1. Common connectivity changes supporting learning across modes 109
5.5.2. Condition-specific connectivity changes supporting learning 110
5.5.3. Learning-related connectivity changes associated with better accuracy 114
5.6. CONCLUSIONS 116
5.7. SUPPLEMENTARY TABLES 117
5.7.1. Network Labels 123
6. STUDY 3: ONLINE 10HZ RTMS DURING LEARNING VIA INSTRUCTION AND EXPLORATION 124
6.1. INTRODUCTION 124
6.2. METHOD 126
6.2.1. Experimental manipulations and TMS paradigm 127
6.2.2. TMS procedure 130
6.3. ANALYSIS AND RESULTS 131
6.3.1. TE learning trials 132
6.3.2. Implementation trials 134
6.3.3. Questionnaire 137
6.4. DISCUSSION 137
6.4.1. Replication of prior findings 138
6.4.2. Accuracy- and RT-specific TMS Effects 138
6.4.3. Non-TMS-Specific Effects 138
6.4.4. Possible Explanations for rAG Effects 139
6.4.5. Possible Mechanisms: Attention and Rule Stabilization 139
6.4.6. Non-Linearity in rAG Activity and TMS Effects 140
6.4.7. The rMFG group 140
6.4.8. Limitations, future directions and conclusions 141
7. GENERAL DISCUSSION 142
7.1. THE ROLE OF THE STRIATUM AND FRONTOSTRIATAL CONNECTIVITY IN INSTRUCTED LEARNING 144
7.1.1. Brain flexibility and temporal sensitivity in Instruction-based learning 145
7.1.2. Ventral striatum in instruction: Intrinsic reward or saliency? 148
7.2. COVERT PRACTICE AND LEARNING PERFORMANCE: INSIGHTS FROM OBSERVATION-BASED LEARNING 150
7.3. THE CEREBELLUM IN REINFORCEMENT LEARNING 151
8. GENERAL CONCLUSIONS: TEMPORAL SENSITIVITY OF THE LEARNING PROCESS 154
9. BIBLIOGRAPHY 157
10. APPENDIX 174
ERKLÄRUNG GEMÄß § 5 DER PROMOTIONSORDNUNG 182
Energy demand of German households and saving potential
No full textThe implementation of the principles of sustainable development requires both using potentialities in saving resources and cutting down emissions (efficiency strategies) as well as more conscious patterns of
behaviour of the actors involved (sufficiency strategies). Starting from the current situation of annual CO2 emissions of about 10 t and a sustainability goal of 1–2 t CO2 emissions per inhabitant and year, the question arises in how far households can contribute to achieve this goal. Therefore, in this paper, the environmental impacts of the energy demand of German households will be evaluated by means of describing its status quo and there from deriving saving potentials
Models and Predictions in Computational Hematology
No full textDie zunehmende Verfügbarkeit experimenteller Daten und Fortschritte in Rechenmethoden haben die Anwendung computergestützter Verfahren in den Lebenswissenschaften stark gefördert, insbesondere in der Hämatologie. Verschiedene Datentypen wie molekulare Profile, fluoreszenzaktivierte Zellsortierung (FACS), Bild- und klinische Daten bieten umfangreiche Einblicke in Krankheitsmechanismen. Die Verknüpfung dieser heterogenen Datenquellen ermöglicht ein tieferes Verständnis hämatologischer Erkrankungen.
Ein zentraler Fokus dieser Arbeit liegt auf der Analyse von Zeitreihendaten, um Krankheitsverläufe zu modellieren und Therapieansätze zu verbessern. Diese Methoden ermöglichen Vorhersagen, z. B. über Rückfallwahrscheinlichkeiten bei Leukämiepatienten nach Absetzen der Medikation oder die Wirksamkeit personalisierter Behandlungen. Zeitreihenanalysen und mathematische Modelle werden genutzt, um dynamische biologische Systeme auf molekularer und zellulärer Ebene zu untersuchen.
Die Dissertation behandelt zwei Hauptthemen:
1. Die Entwicklung und Anwendung neuer Methoden zur Analyse von Zeitreihendaten, illustriert durch zwei medizinische Beispiele: Chronische myeloische Leukämie (CML) und Gentherapie. In der CML-Forschung wurden Modelle entwickelt, die patientenspezifische Merkmale berücksichtigen und Unterschiede in der Wirkung von Medikamenten wie Imatinib und Dasatinib aufzeigen. Simulationen personalisierter Dosierungen demonstrieren das Potenzial für verbesserte Behandlungsergebnisse. Im Bereich der Gentherapie wurde eine neue Metrik eingeführt, die eine frühzeitige Erkennung pathologischer Entwicklungen ermöglicht. Zudem wird die Rolle der Datenqualität bei der Modellgenauigkeit beleuchtet.
2. Die Optimierung interdisziplinärer Zusammenarbeit zwischen Medizinern, Biologen und Modellierern. Hierfür wurde ein benutzerfreundliches Web-Tool entwickelt, das den Zugang zu Rechenmodellen erleichtert, deren Anwendung durch Nicht-Theoretiker unterstützt und die Integration modellbasierter Erkenntnisse in klinische Entscheidungsprozesse fördert.
Methodisch kombiniert die Arbeit agentenbasierte und analytische Ansätze, die an Patientendaten angepasst wurden. Optimierungstechniken spielen eine Schlüsselrolle bei der Modellanpassung. Technologische Lösungen wie Webtechnologien werden eingesetzt, um die Verbreitung von Modellen zu erleichtern und die interdisziplinäre Zusammenarbeit zu fördern.
Die Ergebnisse zeigen, dass mathematische Modelle effektive Werkzeuge zur Verbesserung der personalisierten Medizin sind. In der CML-Forschung wurden Modelle entwickelt, die Unterschiede in der Medikamentenwirkung identifizieren und die Bedeutung personalisierter Dosierungen unterstreichen. Im Bereich der Gentherapie wurde die Bedeutung der Datenqualität für die Genauigkeit von Vorhersagen hervorgehoben. Das Web-Tool dient als Plattform, um Rechenmodelle unabhängig von Implementierungsdetails zu teilen, und unterstützt deren Einbettung in den klinischen Alltag.
Die Dissertation zeigt, dass interdisziplinäre Zusammenarbeit essenziell ist, um komplexe medizinische Fragestellungen zu beantworten. Gleichzeitig bestehen weiterhin Herausforderungen wie Kommunikationsbarrieren, unterschiedliche Terminologien und rechtliche Hürden. Die Arbeit liefert wertvolle Beiträge, um diese Hürden zu überwinden und die Forschung in den Lebenswissenschaften voranzutreiben.
Insgesamt unterstreicht die Arbeit das Potenzial computergestützter Methoden für die Hämatologie und personalisierte Medizin. Die vorgestellten Ansätze bieten eine Grundlage für verbesserte klinische Entscheidungen und die Entwicklung präziserer Behandlungsmethoden.The increasing availability of experimental data and advancements in computational methods have significantly enhanced the application of computer-based approaches in the life sciences, particularly in hematology. Various data types, such as molecular profiles, fluorescence-activated cell sorting (FACS), imaging, and clinical data, provide valuable insights into disease mechanisms. Integrating these heterogeneous data sources enables a deeper understanding of hematological disorders.
A central focus of this work is the analysis of time-series data to model disease progression and improve therapeutic strategies. These methods allow for predictions, such as the likelihood of relapse in leukemia patients after discontinuing medication or the efficacy of personalized treatments. Time-series analyses and mathematical models are utilized to investigate dynamic biological systems at molecular and cellular levels.
The dissertation addresses two main themes:
1. The development and application of novel methods for time-series data analysis, illustrated through two medical examples: chronic myeloid leukemia (CML) and gene therapy. In CML research, models were developed to account for patient-specific characteristics, highlighting differences in the effects of drugs such as imatinib and dasatinib. Simulations of personalized dosages demonstrate the potential for improved treatment outcomes. In the field of gene therapy, a new metric was introduced to enable the early detection of pathological developments. The importance of data quality for prediction accuracy is also emphasized.
2. Optimizing interdisciplinary collaboration between medical practitioners, biologists, and computational modelers. To this end, a user-friendly web tool was developed to facilitate access to computational models, support their use by non-theorists, and integrate model-based insights into clinical decision-making processes.
Methodologically, the work combines agent-based and analytical approaches, which are tailored to patient data. Optimization techniques play a key role in model adaptation. Technological solutions, such as web technologies, are employed to disseminate models and enhance interdisciplinary collaboration.
The results demonstrate that mathematical models are effective tools for advancing personalized medicine. In CML research, models were created to identify differences in drug effects and underscore the importance of personalized dosages. In the context of gene therapy, the significance of data quality for prediction accuracy was highlighted. The web tool serves as a platform for sharing computational models independently of implementation details and supports their integration into clinical practice.
The dissertation highlights the essential role of interdisciplinary collaboration in addressing complex medical questions. However, challenges such as communication barriers, differing terminologies, and legal obstacles remain. This work provides valuable contributions to overcoming these hurdles and advancing research in the life sciences.
Overall, the dissertation underscores the potential of computational methods for hematology and personalized medicine. The presented approaches lay the foundation for improved clinical decision-making and the development of more precise treatment methods
Educational Suicide Prevention in Secondary Schools: The Evaluation of the HEYLiFE Prevention Program
No full textIn the past decade, adolescents' mental well-being has significantly deteriorated, and suicide is the second-leading cause of death in numerous nations. Strategies to improve mental health and to prevent suicide in this age group are called for. School-based, educational prevention programs are among the most promising strategies for suicide prevention with adolescents. However, the research on their effectiveness and safety has encountered several challenges. Up to date, there is no shared theoretical model to guide program development and evaluation. Furthermore, the nature of the phenomenon of suicidality (relative rarity, lethality) poses difficulties in designing methodologically sound evaluation studies. Data on the differential effects of programs on different groups of young people are lacking. The Network for Suicide Prevention in Dresden (NeSuD) project was initiated to improve suicide prevention in the city of Dresden (Saxony, Germany). After a literature review on educational suicide prevention, this thesis presents the results of the evaluation of HEYLiFE, an educative suicide prevention program for secondary schools developed during this project.
First, this thesis aimed to contribute to the theoretical foundation of educational suicide prevention. A three-staged Delphi survey was conducted to explore important contents, target outcomes, and methods of effective and safe educational programs. The Delphi survey is a method that allows to assess expert opinions reliably, reducing unfavourable group decision processes. Participating experts suggested that, above reducing suicidal ideation and attempts, educational programs should aim to increase help-seeking intentions and behaviour, enhance the quality of social support between peers, improve mental health literacy and life-skills such as coping with stress, emotional regulation and problem solving. The experts also proposed to embed suicide prevention in educational programs with a larger scope, to facilitate help-seeking among the program participants and to establish suicide prevention measures on multiple levels in schools to enhance the safety of educational programs. The Delphi survey served as a significant base for researchers to develop school-based, educational suicide prevention programs.
Enhancing mental health literacy regarding depression and suicide has been one of the central strategies for educational programs in the last decades. However, the construct of mental health literacy has been discussed critically due to an unsharp definition and unreliable assessments. These problems caused study results to be tautological and confounded. To solve this problem, it is crucial to define and explore what constructs play a role in enabling people to achieve and maintain a good mental health. Although mental health knowledge is considered the core component of mental health literacy, there is a lack of comprehensive reviews on how it interacts with other constructs relevant to prevention. Through a systematic review, we explored the correlation between mental health knowledge, mental health-related stigma and help-seeking. The review showed that mental health knowledge had a medium-sized, negative correlation with personal stigma (Mdn r = -.28) and a medium-sized positive correlation with attitudes towards help-seeking (Mdn r = .29). The correlations to self-stigma (Mdn r = -.18), help-seeking intention (Mdn r = .15) and help-seeking behaviour (Mdn r = .15) were low. Public stigma was not consistently related to mental health knowledge. These findings contributed to a more profound understanding of the construct of mental health knowledge.
HEYLiFE, an educational suicide prevention program for secondary schools, was a key component of the NeSuD project. HEYLiFE’s efficacy, safety, and acceptability were evaluated with a RCT with waiting-control-group with 745 secondary-school-students aged 12 or older. Outcomes were measured immediately after the intervention (short term) and after 6 months (mid term). We used linear mixed models (LMM) for analysing ordinal outcomes and generalized linear mixed models (GLMM) for binary outcomes, controlling for the nested nature of the data. The program led to an improvement of mental health knowledge and favourable attitudes towards suicidality in the short term. In the mid term, the intervention group showed a more favourable development for help-seeking intentions and risk-factors for suicidality (hopelessness, isolation, burdensomeness, sense of entrapment) from baseline to follow-up than the control group. The program had mixed effects on stigma, with a paradoxical increase in stigma (lower prosocial emotional reaction, higher wish for social distance) in the short term but a more favourable development in the intervention group regarding social distance at follow-up. The evaluation study suggested that the program was acceptable and safe. HEYLiFE is a promising intervention for suicide prevention among adolescents and young adults.
Moreover, this dissertation examines differential effects of HEYLiFE for gender, age, and risk status for suicide attempts. In the main evaluation study (N = 745), HEYliFE had less favourable effects for males than for females on stigmatizing emotional responses to a suicidal peer in the short term and on wish for social distance and help-seeking behaviours in the mid term. Participants in the youngest age group (12-13 years) gained more knowledge than the older ones (14-16; 17+ years) in the mid term. However, they seemed to have stigmatizing emotional reactions to suicidal peers in the short term and did not profit as much in the mid term in terms of a reduction of risk factors. In a second study (N = 218, 14-18 years old), three suicide risk clusters were built using cluster analysis based on suicidality, depression, impulsivity/carelessness and emotional avoidance. We assessed the short time effects on suicide knowledge, agency, and help seeking intentions in case of suicidal thoughts. While knowledge improved in all groups, agency, and help seeking intentions only improved in the low and middle risk group. This study confirmed that HEYLiFE is an effective intervention for suicide prevention. Adolescents already suffering from suicidal ideation and behaviour, however, should receive more targeted interventions.
In summary, this thesis contributes to our understanding of school-based, educational suicide prevention and supports the use of interventions such as HEYLiFE to prevent suicidality among adolescents. Future research should explore how these interventions can be tailored for the needs of males, younger adolescents and adolescents at risk.:Dedication iii
Acknowledgements iv
Statement on the Included Publications vi
Summary viii
Contents xi
List of Figures xiv
List of Tables xv
List of Abbreviations xvi
Chapter 1 Introduction 18
1.1 Suicide Among Adolescents: A Global Health Issue 18
1.2 Suicide Prevention: From General Methods to School-Based Practice 20
1.3 Ongoing Challenges for Educational Suicide Prevention 22
Chapter 2 Literature Review 27
2.1 Suicidality: Definition and Terms 27
2.2 Suicide Among Adolescents and Young Adults 29
2.2.1 Trajectories of Suicidality 30
2.2.2 Risk and Protective Factors 31
2.2.3 Demographic Aspects 34
2.3 Psychological Theories of Suicide 37
2.3.1 Hopelessness Theory of Suicide 38
2.3.2 Interpersonal Theory of Suicide and Ideation-to-Action Framework 39
2.3.3 Integrated Motivational-Volitional Model of Suicide 41
2.4 Educational Suicide Prevention 43
2.4.1 Goals and Strategies of Educational Programs 45
2.4.2 Educational Prevention Program Evaluation 52
2.5 Development of the Research Questions 58
2.5.1 Theoretical Framework: Research Questions 1.1 & 1.2 58
2.5.2 Program Development and Evaluation: Research Question 2 59
2.5.3 Differential Results: Research Questions 3.1 & 3.2 59
Chapter 3 Dos and Don'ts in Designing School-Based Awareness Programs for Suicide Prevention 61
Chapter 4 Beyond Knowledge: A Systematic Review on the Correlation of Mental Health Literacy with Stigma and Help Seeking 72
Chapter 5 Addressing Help-seeking, Stigma and Risk Factors for Suicidality in Secondary Schools: Short-term and Mid-term Effects of the HEYLiFE Suicide Prevention Program in a Randomized Controlled Trial 95
Chapter 6 Promoting Protective Factors for Suicidal Behavior in Adolescents at Risk: Differential Efficacy of the HEYLiFE Suicide Prevention Program 122
Chapter 7 Discussion 143
7.1 Theoretical Framework 143
7.1.1 Expert Opinions? It’s a Start 143
7.1.2 School-based Prevention: A Worthwhile Commitment 145
7.1.3 Heading to Shared Theoretical Frameworks for Suicide Prevention 146
7.1.4 From Specific to Upstream Prevention 147
7.1.5 Enhancing Safety 147
7.1.6 Mental Health Knowledge: The Core of Mental Health Literacy 149
7.1.7 Quantifying Correlations: How Strong is the Link between Knowledge, Stigma and Help-Seeking? 150
7.1.8 Unravelling the Correlations: How does Knowledge Contribute to Prevention? 152
7.1.9 The Limits of Knowledge for Suicide Prevention 154
7.2 Program Development and Evaluation 155
7.2.1 Short-lived Effects for Knowledge and Attitudes 156
7.2.2 Protective Effects on the Mid-term 158
7.2.3 Heterogeneous Findings for Stigma 160
7.2.4 Program Safety 162
7.2.5 Program Acceptability 163
7.3 Differential Effects 164
7.3.1 Gender: Masculinity Norms as a Challenge for Prevention 164
7.3.2 Age: Tailoring Prevention to Development 167
7.3.3 Suicide Risk: When Educational Prevention is not Enough 168
7.4 Limitations 170
7.5 Implications for Research and Practice 173
7.5.1 Specific vs. Upstream Prevention 173
7.5.2 Central Mechanisms and Outcomes in Specific Educational Suicide Prevention 174
7.5.3 Extension, Replicability and Generalizability of the Results 175
7.5.4 Further Understanding Differential Effects 176
7.5.5 Towards Multi-Level Suicide Prevention in Schools 177
7.5.6 Lessons Learned During the NeSuD Project 178
7.5.7 Outlook: Recent Developments of the NeSuD Project 180
7.6 Conclusion 181
References 183
Appendix A: Additional Materials Grosselli et al., 2021 236
Appendix B: Additional Materials Grosselli et al., 2024b 246
Appendix C: Additional Materials Grosselli et al., 2024a 251
Erklärung gemäß § 5 der Promotionsordnung 26
Machine Learning Algorithms for Epileptic Seizure Prediction
No full textMit etwa 1 % der Weltbevölkerung ist die Epilepsie eine der häufigsten neurologischen Erkrankungen. Bei einem großen Teil der Patienten können die Anfälle mit Medikamenten nicht ausreichend kontrolliert werden. Die Entwicklung eines Geräts zur Vorhersage von Anfällen birgt das Potenzial, die Angst und Ungewissheit im Zusammenhang mit dem Auftreten von Anfällen zu verringern, das Verständnis für die Krankheit zu verbessern und bei Bedarf die sofortige Verabreichung von Maßnahmen zu erleichtern. In dieser Arbeit werden verschiedene überwachte und unüberwachte Algorithmen des maschinellen Lernens für die Vorhersage epileptischer Anfälle auf der Grundlage des intrakraniellen Elektroenzephalogramms entwickelt und evaluiert. Anhand nicht-kontinuierlichen Benchmark-Daten wird gezeigt, dass vollständig datengetriebene Methoden aus dem Bereich des Deep Learning den Stand der Technik erreichen können. Darüber hinaus zeigen grundlegend unterschiedliche Algorithmen eine ähnliche Leistung sowie Kohärenz in den fehlerhaften Stichproben. Post-hoc-Experimente deuten zudem darauf hin, dass weitere Verbesserungen auf diesem Datensatz begrenzt sein könnten. Analysen von kontinuierlichen Daten zeigen das Vorhandensein einer zyklischen und/oder gerichteten zeitlichen Entwicklung in den Verteilungen der EEG-Daten für alle Patienten. Unüberwachte Methoden zeigen eine klinisch relevante Leistung nur für Patienten, die eine zyklische zeitliche Entwicklung in Kombination mit phasengleichem Auftreten von Anfällen relativ zu dieser Entwicklung aufweisen. Darüber hinaus wird gezeigt, dass die Einbeziehung aktueller Daten in den Trainingssatz zu einer verbesserten Leistung des Klassifikators führt, während die Einbeziehung älterer Daten nur marginale Verbesserungen mit sich bringt. Das deutet darauf hin, dass die Klassifizierung in erster Linie auf aktuellen Daten beruht. Schließlich macht die Studie deutlich, dass die mit Benchmark-Daten erzielte Leistung möglicherweise zu optimistisch ist. Die Definitionen, die zum Labeln dieser Daten verwendet wurden, könnten den klinischen Nutzen des Klassifikators möglicherweise einschränken. Ein klinisch relevantes, vollständig prospektives System zur Vorhersage von Anfällen, das auf dem vorgeschlagenen Klassifikator basiert, wurde getestet und schnitt nicht signifikant besser ab als ein zufälliger Prädiktor. Dies deutet auf die Herausforderungen hin, robuste Vorhersagen in einer realen Anwendung zu erzielen.With approximately 1% of the global population affected, epilepsy is one of the most prevalent neurological diseases. For a substantial portion of patients, medication can not control seizures sufficiently, limiting their treatment options. The development of a predictive device for seizures holds the potential to alleviate the anxiety and uncertainty surrounding seizure occurrences, enhance disease understanding, and facilitate prompt administration of interventions when needed. This thesis designs and evaluates various supervised and unsupervised machine learning algorithms for the prediction of epileptic seizures based on intracranial Electroencephalogram. It demonstrates, that fully data-driven methods from the field of deep learning can achieve state-of-the art performance. Moreover, fundamentally different algorithms exhibit similar performance as well as coherence in the erroneous segments when evaluated on non-continuous benchmark data. Furthermore, post-hoc experiments suggest that further improvement on this dataset may be limited. Analysis of continuous data reveals the presence of cyclic and/or directed temporal evolution in the distributions of EEG data for all patients. Unsupervised methods demonstrate clinically relevant performance only for patients exhibiting cyclic temporal evolution in combination with phase-locked seizure occurrences relative to this evolution. Additionally, the study shows that incorporating recent data into the training set leads to increased classification performance, while the inclusion of older data yields only marginal improvements, suggesting that classification primarily relies on recent data. Lastly, the study highlights that performance achieved on benchmark data may be overly optimistic. The definitions that have been used for labelling this data could potentially limit the clinical utility of the classifier. A clinically relevant, fully prospective seizure prediction system based on the proposed classifier is tested and found to perform not significantly better than a random predictor, indicating the challenges of achieving robust predictions in a real-world scenario
Learning by Doing: Insights from Power Market Modelling in Energy Economics Courses
No full textMuch of energy economics curricula involves the study of techno-economic aspects of energy systems with an increasing focus devoted to fostering an understanding of the interactions between innovative technologies and adaptive markets. As the interplay of these dynamics and their impacts on market equilibria and outcomes is quite complex, optimization models are well-suited to facilitate their study. This paper presents two exemplary model approaches and associated case studies, which can be employed to study market developments driving long-term adaptations in the portfolio of power-generation assets as well as scheduling problems of individual plant owners with a focus on assessing the impact of changing market conditions on the profitability of investments. The combination of these two modelling approaches constitutes an innovative means of facilitating students’ understanding of how individual decisions of different market stakeholders lead to welfare-maximizing market equilibria under the assumption of perfect competition. The models are discussed along with the experiences acquired employing them in various forms as project assignments. In summary, the integration of modelling exercises and assignments into the curriculum of energy economics courses has proven to be a practical means of reinforcing and broadening lecture material that is both interesting and rewarding for students
Digitale Vermittlung in geisteswissenschaftlichen Fächern an Schulen: Zwischenstände und Bedarfsanalyse
No full textDigital Education: Competence DevelopementAus dem Text: Die Begriffe „digitale Lehre“ und „Vermittlung von Digitalkompetenzen“ sind nicht erst seit der COVID-19-Pandemie 2020 präsente Themen der Bildungswissenschaften und Bildungspolitik. Bereits 2016 wurde im Abschlussbericht der deutschen Kultusministerkonferenz definiert, dass Lehrkräfte zusätzlich zu ihrem Fachbereich über ein Mindestmaß an allgemeiner Medienkompetenz verfügen müssen, insbesondere im Hinblick auf die wachsende Relevanz einer digitalen Bildung (KMK, 2016, S. 24–25). Der DigitalPakt von 2019, dessen Erneuerung in diesem Jahr von der Bundesregierung besprochen wird, greift diese Forderung auf, setzt die Schwerpunkte allerdings auf die digitale Infrastruktur von Schulen und die technischen Möglichkeiten (Digitalpakt Schule, 2019, S. 1–3)
Workshop: Generative KI und akademische Workshop Integrität – ein Widerspruch?
No full textDigital Education: Hands onAus dem Text: Die Nutzung von Generativer KI (García-Peñalvo & Vázquez-Ingelmo, 2023) am Hochschulsektor wird mittlerweile in zahlreichen wissenschaftlichen Publikationen diskutiert, wobei Chancen ebenso thematisiert werden wie Herausforderungen und Barrieren (Michel-Villarreal et al., 2023). Die mit Generativer KI verbundenen umfassenden Veränderungen in nahezu allen Bereichen der akademischen Welt wird dabei sowohl positiv als auch kritisch gesehen. Davon betroffen ist auch die akademische Integrität. Mithilfe textgenerierender KI können mittlerweile einzelne Absätze bis hin zu ganzen Essays anhand einfacher
Prompts generiert werden. Ob bzw. unter welchen Bedingungen diese Texte als eigenständige Leistung deklariert werden können, ist derzeit Gegenstand vieler Debatten. Fest steht jedoch, dass derart generierte Texte aktuell weder durch Personen noch mittels Softwaresystemen treffsicher zu detektieren sind (Alexander et al., 2023). Davon ausgehend, dass derzeitige Mängel der KI wie z.B. halluzinierte Inhalte oder das Erfinden von Literaturangaben in Zukunft größtenteils behoben werden, könnte man annehmen, dass Generative KI tatsächlich eine Bedrohung für die Grundprinzipien der akademischen Integrität darstell