119092 research outputs found
Sort by
WGP-Standpunkt Energieeffizienz, Energieflexibilität und Dekarbonisierung
No full textDie WGP (Wissenschaftliche Gesellschaft für Produktionstechnik e.V.) ist ein Zusammenschluss führender deutscher Professorinnen und Professoren der Produktionswissenschaft. Sie vertritt die Be-lange von Forschung und Lehre gegenüber Politik, Wirtschaft und Öffentlichkeit. Die WGP vereinigt 64 Professorinnen und Professoren aus knapp 40 Universitäts- und Fraunhofer-Instituten und steht für rund 2.000 Wissenschaftlerinnen und Wissenschaftler der Produkti-onstechnik. Die Mitglieder genießen sowohl in der deutschen Wis-senschaftslandschaft als auch international eine hohe Reputation und sind weltweit vernetzt. Die Labore der Mitglieder sind auf einem hohen technischen Stand und erlauben den Professorinnen und Pro-fessoren der WGP, in ihren jeweiligen Themenfeldern sowohl Spit-zenforschung als auch praxisorientierte Lehre zu betreiben. Die WGP hat sich zum Ziel gesetzt, die Bedeutung der Produktion und der Produktionswissenschaft für die Gesellschaft und für den Standort Deutschland aufzuzeigen. Sie bezieht Stellung zu gesellschaftlich re-levanten Themen von Industrie 4.0 über Energieeffizienz bis hin zu 3D-Druck
Identifying grain boundary and intragranular pinning centres in Sm2(Co,Fe,Cu,Zr)17 permanent magnets to guide performance optimisation
No full textPermanent magnets draw their properties from a complex interplay of chemical composition and phase, each with their associated intrinsic magnetic properties. Gaining an understanding of these interactions is the key to deciphering the origins of a permanent magnets’ magnetic performance and facilitate the engineering of much improved-performing magnets. Here, we use advanced multiscale microscopy and microanalysis on a bulk Sm 2 (Co,Fe,Cu,Zr) 17 pinning-type high-performance magnet with outstanding thermal and chemical stability. Comparison of the microstructure in regions of different composition, we demonstrate that the pinning of magnetic domains, imaged by nanoscale magnetic induction mapping, is controlled by the composition and atomic arrangement of copper. This is confirmed by micromagnetic simulations. Contrary to the belief that grain boundaries are “weak links” in magnetic materials, we demonstrate grain boundaries undergo magnetization reversal at relatively low fields (0.1-0.3 T), but this remains confined to these regions and does not significantly impact the magnet’s coercivity. Our results showcase that it is the optimal microstructure within the grain itself that is crucial for achieving the desired magnetic properties
Gas evolution in Ruddlesden–Popper-type intercalation cathodes in all-solid-state fluoride-ion-batteries: implications on battery performance and synthesis of highly oxidized oxyfluorides
No full textIn this work, we report the gas evolution behavior of two common intercalation-based cathode materials for fluoride-ion-batteries (La2NiO4 and La2CoO4) by chemical fluorination and electrochemical fluorination, as well as the thermal stability of the electrochemically fluorinated active materials. Variable-temperature X-ray diffraction with coupled mass spectrometry shows that the fluorinated phases of both active materials release oxygen after oxidative fluorination using AgF. Differential electrochemical mass spectrometry revealed that significant CO2, CO and NO evolution takes place during electrochemical fluorination, but O2 evolution could not be detected. Consequently, the gas evolution observed does not compromise the long-term stability of the cell performance. Variable-temperature X-ray diffraction of the charged cathode composites reveals that the electrochemically fluorinated products decompose once heated beyond 300 °C, and then undergo similar phase evolution behavior as observed in chemical fluorination. Thus, we conclude that oxygen evolution in RP-type materials on oxidative fluorination requires thermal activation beyond the operational temperature of 170 °C, and that it can be kinetically suppressed regardless of the reduced Fermi level induced on fluorination. Our findings indicate that concepts used for the oxidative topochemical fluorination of materials need to be reconsidered: the choice of fluorination conditions must be carefully adopted to the changes of electronic structure induced and the temperature dependence of the oxygen evolution of the highly oxidized state
Rejuvenating IRS: AoI-Based Low Overhead Reconfiguration Design
No full textIntelligent reflective surface (IRS) technologies help mitigate undesirable effects in wireless links by steering the communication signal between transmitters and receivers. IRS elements are configured to adjust the phase of the reflected signal for a user’s location and enhance the perceived signal-to-noise ratio (SNR). In this way, an IRS improves the communication link but inevitably introduces more communication overhead. This occurs especially in mobile scenarios, where the user’s position must be frequently estimated to re-adjust the IRS elements periodically. Such an operation requires balancing the amount of training versus the data time slots to optimize the communication performance in the link. Aiming to study this balance with the age of information (AoI) framework, we address the question of how often an IRS needs to be updated with the lowest possible overhead and the maximum of freshness of information. We derive the corresponding analytical solution for a mobile scenario, where the transmitter is static and the mobile user (MU) follows a random waypoint mobility model. We provide a closed-form expression for the average peak age of information (PAoI), as a metric to evaluate the impact of the IRS update frequency. As for the performance evaluation, we consider a realistic scenario following the IEEE 802.11ad standard, targeting the mmWave band. Our results reveal that the minimum achievable average PAoI is in the microsecond range and the optimal IRS update period is in the seconds range, causing 9% overhead in the link when the MU moves at a velocity of 1 m/s
Advancing Learned Cost Models For Modern Data Systems
No full textData systems form the backbone of every modern IT architecture by storing and processing vast amounts of data. Depending on the workload, the users, and their performance and scalability requirements, modern data systems typically need to be adapted and optimized for high performance, which is a challenging task. However, the way of adapting and optimizing data systems has changed fundamentally in recent years, driven by the megatrend of cloud computing. While companies used to host and optimize their data systems themselves, data systems such as Database Management Systems (DBMSs) or Distributed Stream Processing Systems (DSPSs) are now increasingly being deployed in cloud environments. This trend shifts the task of system adaptation and optimization to the cloud providers, but in fact makes it even more challenging due to the vast diversity of customers and workloads that need to be considered.
At the same time, the trend towards the cloud is a unique opportunity, as it enables cloud providers to collect data from their wide variety of customers, enabling the use of machine learning to adapt and improve data systems automatically. The concept of learned database components was introduced to either optimize or fully replace traditional data system components with the help of machine learning. At the core of this idea, learned database components are trained on previous customer workload traces and, by that, are expected to achieve higher performance than traditional components.
In this thesis, we focus on cost models that serve as a core building block for various system optimizations, such as query optimization, workload and query scheduling or resource admission. Recently, Learned Cost Models (LCMs) were proposed that predict execution costs of a specific database query or a system configuration with the help of machine learning. A core requirement of LCMs is to make precise predictions even for queries or databases that differ significantly from their training data, as queries and databases in real-world data systems constantly evolve. For this reason, recent research on LCMs has primarily focused on two main aspects: improving their accuracy and increasing their generalizability. As a consequence, LCMs were able to provide highly accurate predictions and outperformed traditional approaches substantially – even for scenarios that substantially differ from the training data.
However, in this thesis, we argue that the potential of LCMs remains underexploited and that significant limitations persist with respect to their functionality and evaluation, which we outline below. Critically, these limitations significantly hinder their practical applicability for various optimization tasks. For this reason, we provide four contributions to overcome existing limitations of LCMs and, by that, advance LCMs for modern data systems.
First, existing LCMs are unable to solve various optimization tasks, such as the operator placement on heterogeneous hardware, which is, however, important in distributed systems, IoT, and streaming scenarios. The reason is that they do not yet take hardware and network resources into account, which affect execution costs significantly. Thus, in the first contribution of this dissertation, we first extend LCMs towards Distributed Stream Processing System (DSPS), and then towards the consideration of heterogeneous hardware resources. We propose COSTREAM, a novel LCM that takes hardware and network resources into account and, by that, enables the optimized placement of query operators to the hardware plane. By precisely estimating query execution costs for DSPS, COSTREAM is able to identify optimized initial operator placements that come with low processing latencies and, by that, improves system performance substantially by providing speed-ups of up to 21×.
Second, existing LCMs are unable to take user-defined functions (UDFs) into account. UDFs enable users to express complex or domain-specific computational logic and are increasingly used to analyze and process data within a DBMS. However, as UDFs contain arbitrary user code, query optimizers fail to estimate their effects on a query plan. As a result, they treat UDFs as black boxes and rely on static optimization rules, which often leads to significant performance degradation. Therefore, in the second contribution of this dissertation, we propose to extend LCMs for UDFs. We introduce GRACEFUL, a novel LCM that considers UDFs during cost estimation by leveraging Graph Neural Networks (GNN). Thus, it allows an informed decision about the positioning of UDFs in the physical query plan. Hence, the use of GRACEFUL leads to substantial performance improvements of queries with UDFs, such as speed-ups of up to 50×.
Third, the extent to which the fundamental task of query optimization in DBMSs benefits from the high accuracy of LCMs in terms of system performance has not yet been investigated. Therefore, in the third contribution of this dissertation, we present a comprehensive evaluation study that analyzes seven current LCMs and evaluates in detail how good they really are for query optimization. As a central capability, we investigate how well LCMs are able to select the optimal plan from different candidates and rank plan candidates correctly in fundamental tasks such as join ordering. Our findings indicate that although LCMs are generally capable of learning complex cost functions; they are often still inferior to traditional approaches when selecting optimal execution plans. Therefore, to improve future LCMs, we present recommendations and novel evaluation metrics to improve future LCMs for query optimization.
Finally, given that LCMs are usually based on deep learning architectures, they are typically inexplicable in their prediction and thus exhibit so-called black-box behavior. As a result, it remains unclear how and why exactly LCMs come to a specific prediction, which fundamentally hinders targeted debugging and troubleshooting. Thus, in our fourth contribution, we aim to make LCMs debuggable by explaining their predictions and visualizing what they have really learned. For this, we provide methods, initial results and an interactive demo that facilitate explaining the predictions of LCMs.
Overall, this dissertation presents novel methods and approaches that advance LCMs, address their existing limitations, and unlock their full potential. To enhance their functionality, we propose incorporating heterogeneous hardware resources and UDFs into cost prediction, which demonstrates substantial performance benefits for advanced query optimization problems. To improve the evaluation of LCMs, we conduct a detailed study and introduce novel evaluation metrics and practical recommendations towards making LCMs more applicable for query optimization. Moreover, we apply explainability methods to make the predictions of LCMs more understandable and to enable systematic troubleshooting. Finally, we present and discuss future research directions enabled by this thesis to advance LCMs for modern data systems
On the Development of AI in Germany
No full textThe article gives a brief account of the historical evolution of Artificial Intelligence in Germany, covering key steps from antiquity to the present state of the discipline. Its focus is on AI as a science and on organisational aspects rather than on technological ones or on specific AI subjects
Herold, Emanuel: Utopien in utopiefernen Zeiten. Zukunftsdiskurse am Ende der fortschrittlichen Moderne, 280 S., Wallstein, Göttingen 2020.
No full textWie kann sich in Zeiten von Krise, Krieg und Pandemie eine erstrebenswerte Zukunft vorgestellt werden? Ist das überhaupt zeitgemäß? Emanuel Herold ist sich sicher: jetzt mehr denn je! In seiner umfangreichen Theoriearbeit legt der Autor einen literatursoziologischen Ansatz vor, um die Gattung der Utopie als „Reflexion zeitlicher Verhältnisse“ verständlich zu machen. Der Titel verspricht ein spannendes Buch. Überzeugend ist auch die Einleitung, in der Herold auf die Spannungen zwischen verschiedenen Konstruktionen von Zukunft hinweist. Etwas irritierend wirkt allerdings die zweite Fußnote. Hier heißt es zur Bezeichnung von Berufsgruppen, dass zwar nur die männliche Form genannt werde, die weibliche Form aber stets „mitgemeint“ sei, um den Lesefluss nicht zu beeinträchtigen (S. 10). Bei soziologischen „AutorInnen“ (S. 22) scheint dies jedoch nicht der Fall zu sein
ID-Booth: Identity-consistent Face Generation with Diffusion Models
No full textRecent advances in generative modeling have enabled the generation of high-quality synthetic data that is applicable in a variety of domains, including face recognition. Here, state-of-the-art generative models typically rely on conditioning and fine-tuning of powerful pretrained diffusion models to facilitate the synthesis of realistic images of a desired identity. Yet, these models often do not consider the identity of subjects during training, leading to poor consistency between generated and intended identities. In contrast, methods that employ identity-based training objectives tend to overfit on various aspects of the identity, and in turn, lower the diversity of images that can be generated. To address these issues, we present in this paper a novel generative diffusion-based framework, called ID-Booth. ID-Booth consists of a denoising network responsible for data generation, a variational auto-encoder for mapping images to and from a lower-dimensional latent space and a text encoder that allows for prompt-based control over the generation procedure. The framework utilizes a novel triplet identity training objective and enables identity-consistent image generation while retaining the synthesis capabilities of pretrained diffusion models. Experiments with a state-of-the-art latent diffusion model and diverse prompts reveal that our method facilitates better intra-identity consistency and inter-identity separability than competing methods, while achieving higher image diversity. In turn, the produced data allows for effective augmentation of small-scale datasets and training of betterperforming recognition models in a privacy-preserving manner. The source code for the ID-Booth framework is publicly available at https://github.com/dariant/ID-Booth
Record linkage based patient intersection cardinality for rare disease studies using Mainzelliste and secure multi-party computation
No full textBackground: The low number of patients suffering from any given rare diseases poses a difficult problem for medical research: With the exception of some specialized biobanks and disease registries, potential study participants’ information are disjoint and distributed over many medical institutions. Whenever some of those facilities are in close proximity, a significant overlap of patients can reasonably be expected, further complicating statistical study feasibility assessments and data gathering. Due to the sensitive nature of medical records and identifying data, data transfer and joint computations are often forbidden by law or associated with prohibitive amounts of effort. To alleviate this problem and to support rare disease research, we developed the Mainzelliste Secure EpiLinker (MainSEL) record linkage framework, a secure Multi-Party Computation based application using trusted-third-party-less cryptographic protocols to perform privacy-preserving record linkage with high security guarantees. In this work, we extend MainSEL to allow the record linkage based calculation of the number of common patients between institutions. This allows privacy-preserving statistical feasibility estimations for further analyses and data consolidation. Additionally, we created easy to deploy software packages using microservice containerization and continuous deployment/continuous integration. We performed tests with medical researchers using MainSEL in real-world medical IT environments, using synthetic patient data.
Results: We show that MainSEL achieves practical runtimes, performing 10 000 comparisons in approximately 5 minutes. Our approach proved to be feasible in a wide range of network settings and use cases. The “lessons learned” from the real-world testing show the need to explicitly support and document the usage and deployment for both analysis pipeline integration and researcher driven ad-hoc analysis use cases, thus clarifying the wide applicability of our software. MainSEL is freely available under: https://github.com/medicalinformatics/MainSEL
Conclusions: MainSEL performs well in real-world settings and is a useful tool not only for rare disease research, but medical research in general. It achieves practical runtimes, improved security guarantees compared to existing solutions, and is simple to deploy in strict clinical IT environments. Based on the “lessons learned” from the real-word testing, we hope to enable a wide range of medical researchers to meet their needs and requirements using modern privacy-preserving technologies
Color Matching and Biomimicry for Multi-Material Dental 3D Printing
No full textThe growing global demand for removable partial and full dentures, driven by an aging population and the high prevalence of edentulism, emphasizes the importance of advancing manufacturing solutions. Multi-material jetting, with newly regulatory-approved dental resins, facilitates the production of monolithic, full-color dentures, reducing manual labor and enabling advanced aesthetic customization. We propose a practical method for dental layer biomimicry and multi-spot shade matching using multi-material 3D printing. It integrates seamlessly into workflows combining dental CAD tools, which compute the outer shape of dentures, and industrial multi-material slicers. We introduce a morphable star-shape descriptor to embed enamel, dentin, and root layers with controlled thicknesses into outer tooth geometries. To compute per-layer material ratios, we first present a forward model to predict the color of printed layered slabs
mimicking inner tooth structures with variable translucencies. The slab design enables reasonable color predictions on layered tooth shapes. Based on the forward model, we propose a backward model to compute per-layer material ratios for given layer-translucencies to achieve color matches at multiple points on the tooth surface. The method’s effectiveness is demonstrated by fabricating various dentures with custom layers that accurately replicate VITA classical shades,
showcasing its practical and versatile application in denture manufacturing