KITopen
Not a member yet
63711 research outputs found
Sort by
Improving Label Error Detection and Elimination with Uncertainty Quantification
Get PDFIdentifying and handling label errors can significantly enhance the accuracy of supervised machine learning models. Recent approaches for identifying label errors demonstrate that a low self-confidence of models with respect to a certain label represents a good indicator of an erroneous label. However, latest work has built on softmax probabilities to measure selfconfidence. In this paper, we argue that—as softmax probabilities do not reflect a model’s predictive uncertainty accurately— label error detection requires more sophisticated measures of model uncertainty. Therefore, we develop a range of novel, model-agnostic algorithms for Uncertainty Quantification-Based Label Error Detection (UQ-LED), which combine the techniques of confident learning (CL), Monte Carlo Dropout (MCD), model uncertainty measures (e.g., entropy), and ensemble learning to enhance label error detection. We comprehensively evaluate our algorithms on four image classification benchmark datasets in two stages. In the first stage, we demonstrate that our UQ-LED algorithms outperform state-of-the-art confident learning in identifying label errors. In the second stage, we show that removing all identified errors from the training data based on our approach results in higher accuracies than training on all available labeled data. Importantly, besides our contributions to the detection of label errors, we particularly propose a novel approach to generate realistic, class-dependent label errors synthetically. Overall, our study demonstrates that selectively cleaning datasets with UQ-LED algorithms leads to more accurate classifications than using larger, noisier datasets
Synergistic Two‐Color Photochemical Polymer Network Formation and Lithography
Get PDFWe introduce synergistic two-color lithography as an advanced wavelength-gated strategy for spatially and temporally controlling polymer network formation. Our photoresist entails two photoswitches, i.e., diarylindenone epoxide (DIO) and strained azobenzene (SA), each activated at a judiciously selected wavelength, i.e., 375 or 430 nm. Under specific conditions of photon flux, simultaneous irradiation at both wavelengths induces a (3 + 2) cycloaddition between the photoactivated DIO′ and SA′ species, generating covalently crosslinked networks, whereas under these specifically determined conditions, single-wavelength exposure does not induce solidification. Kinetic analysis highlights the potential of synergistic activation to enable advanced additive manufacturing. We implemented the two-color activated covalent bond forming system in a dual-laser lithographic platform enabling the fabrication of well-defined structures, including segmented ring and butterfly architectures by simply activating and deactivating one of the colors of light
The Next Generation of Planning: A Year of Milestones f rom AESOP Young Academics
Get PDFThe AESOP Young Academics (YA) Network is an integral branch of the Association of European Schools of Planning (AESOP), the leading academic association representing European planning schools. Established in 2003, the YA Network was created to foster the active participation of early-career scholars within the planning discipline and to strengthen their engagement with the broader AESOP community. Open to PhD students, post-doctoral researchers, and academics at the beginning of their academic careers, the YA Network provides an open, inclusive platform for scholarly exchange. At the heart of the network lies the Co-ordination Team (CT ), a group of six elected members who work collectively to initiate, coordinate, and promote YA-related activities throughout their two-year term of service. These activities include the annual YA Conference, the AESOP PhD Work-shop, and active involvement in the AESOP Annual Congress. It played a proactive role in contributing to the success of major AESOP events, including the 36th AESOP Annual Congress in Paris, which brought together over 1200 participants from 55 countries, and the AESOP PhD Workshop in Grenoble, which welcomed 28 doctoral students from around the world. This report offers a comprehensive overview of the YA Network’s activities during the 2024–2025 term, highlighting the continued growth and contribution of young academics to the AESOP community and to the future generation of planning
Contribution of climate change and human activities to streamflow and lake water level variations at regional scales
Get PDFGlobal warming has been intensifying the water cycle, thereby altering regional climate systems and hydrological processes. This is particularly the case for the Poyang Lake Basin (PLB) in monsoon-controlled southeast China, where climate changes and human activities are evident. Our study aims to quantify the contributions of climate change and human activities to the spatiotemporal variations of the relevant variables across meteorological and hydrological compartments on the basin scale. This study applies the moving t-test, Mann–Kendall test, and linear regression models to quantify the impacts of climate change and human activities on changes in streamflow and lake level from 1960 to 2019. Results show that precipitation, streamflow, and air temperature have increased, but Poyang Lake level has declined. Change points in streamflow trends are identified in 1991 and 2002 and in lake level in 2003. Contribution analysis indicates that climate change is the primary driver of increased streamflow. However, after 2002, the contribution of climate change declined, while that of human activities increased. The abrupt decline in lake level is mainly attributed to anthropogenic interventions. These findings identify the dominant factors of hydrological change and provide guidance for ensuring water security and sustainable water resource management in the basin
Experimental investigations of N‐joints made of high‐strength hollow sections
Get PDFThis article presents experimental investigations of the structural behaviour of welded N-joints made of high-strength hollow sections. A total of 13 large-scale tests were carried out on joints made from five different steel grades (S500MH, S620QH, S700MH, S770QH and S890QH) using both circular (CHS) and square (SHS) hollow sections. All specimens have been produced with full penetration butt welds to avoid weld failure and isolate punching shear failure (PSF) and chord face failure (CFF). Welding parameters were optimised based on dilatometer results and thermal cycle limits (t ≤ 12 s). Measurement techniques included strain gauges and digital image correlation (DIC) to quantify joint deformation by chord indentation. The force–indentation relationship was used to determine the ultimate strength of the joints. The results demonstrate the effects of steel grade, β-ratio, chord slenderness and preload on the load-bearing behaviour of the joints and provide a reliable basis for numerical modelling and future design recommendations for N-joints made of high-strength steel
The SFB/TRR 393 Collaborative Research Centre: trajectories of affective disorders – Cognitive–emotional mechanisms of symptom change = Sonderforschungsbereich SFB/TRR 393: Verlaufsformen affektiver Erkrankungen – Kognitiv-emotionale Mechanismen von Symptomveränderungen
No full textMajor depressive disorder (MDD) and bipolar disorder (BD) are prevalent and disabling psychiatric disorders, often following a chronic and relapsing course. The Collaborative Research Centre 393 (SFB/TRR 393), funded by the German Research Foundation (DFG), aims to identify trajectories and symptom changes in MDD and BD, with a focus on cognitive–emotional mechanisms and their neurobiological underpinnings.
Our research initiative seeks to (1) identify individual trajectories of recurrences and remissions in affective disorder (AD), (2) determine cognitive–emotional mechanisms and neurobiological correlates of acute symptom changes, and (3) probe mechanism-based interventions.
These goals will be pursued through a threefold approach: (1) Continuous mobile assessment in a prospective cohort: We will combine in-depth clinical characterization with multilevel neuroimaging, biobanking, and -omics analyses in 1500 AD patients and healthy participants over a 2-year follow-up (German Mental Health Cohort, GEMCO) at three time points. Participants will be drawn from existing DFG FOR 2107 and BMBF Early-BipoLife cohorts (Domain A). (2) Identification of key cognitive-emotional mechanisms: We will study emotion regulation, expectation, social cognition, and cognitive–behavioural rhythms, and their neurobiological correlates mediating symptom changes, using parallel human studies and animal experiments (Domain B). (3) Targeted interventions: We will probe key cognitive–emotional mechanisms in relation to recurrences and remissions (Domain C).
Over a 12-year period, we will elucidate environmental, psychosocial, and (neuro)biological predictors of illness course; cognitive–emotional and neurobehavioural mechanisms underlying real-life recurrences and remissions; and targeted, mechanism-based interventions
The Present and Future of Accountability for AI Systems: A Bibliometric Analysis
No full textArtificial intelligence (AI) systems, particularly generative AI systems, present numerous opportunities for organizations and society. As AI systems become more powerful, ensuring their safe and ethical use necessitates accountability, requiring actors to explain and justify any unintended behavior and outcomes. Recognizing the significance of accountability for AI systems, research from various research disciplines, including information systems (IS), has started investigating the topic. However, accountability for AI systems appears ambiguous across multiple research disciplines. Therefore, we conduct a bibliometric analysis with 5,809 publications to aggregate and synthesize existing research to better understand accountability for AI systems. Our analysis distinguishes IS research, defined by the Web of Science “Computer Science, Information Systems” category, from related non-IS disciplines. This differentiation highlights IS research’s unique socio-technical contribution while ensuring and integrating insights from across the broader academic landscape on accountability for AI systems. Building on these findings, we derive research propositions to lead future research on accountability for AI systems. Finally, we apply these research propositions to the context of generative AI systems and derive a research agenda to guide future research on this emerging topic
Design of a sensor integrating bolt for multiaxial force measurement and development of a design methodology
No full textStructurally integrated force measurements enable real-time monitoring of the overall system status. Existing sensor integration strategies in bolts often compromise mechanical integrity and deviate from standardized external dimensions, limiting their applicability. In this paper, a sensor is presented that is integrated into a M20 bolt to measure axial force and bending torques. The sensor is integrated directly into the bolt itself without altering its external dimensions. This integration method offers a more flexible solution compared to currently available solutions. Furthermore a method is presented that is extracted out of the development process. It includes the optimization of the design space for sensor and electronics as well as a general procedure based on the VDI 2206 and the methodology of testing. The sensor-integrating bolt retains 76.5% of the static load capacity of a standard bolt, which can be compensated by selecting a higher strength class. The sensor body consists of three vertical bending beams with applied strain gauges. The experimental findings have revealed a low linearity error of ±0.82% for axial load and ±0.11% for bending load. Linearity errors were measured at ±1.53% and ±0.52%, respectively. A flexible modular electronic platform has been developed, suited for complete integration into the bolt’s head. It allows experimentation with the sensors read-out electronics, and various modules for communication. The energy supply is evaluated using different energy harvesting and energy transmission concepts. The novel sensor integrating bolt can be used to measure all relevant loads on bolts and to indicate early failure like decrease of clamping force or opening joints. The method can be used to support future sensor integration projects