13979 research outputs found
Sort by
Crystal structure of di[potassium(diaza-18-crown-6)]dicaesium nonastannide( 4–) diethylenediamine, [K(C₁₂H₂₆N₂O₄)]2Cs₂[Sn₉] · 2C₂N₂H₈, a polyanion ² ͚ [KCs₂Sn₉]⁻ with low-dimensional arrangement of [Sn₉] clusters
No full textC₂₈H₆₈Cs₂K₂N₈O₈Sn₉, monoclinic, P121/c1 (No. 14), a = 13.897(3) Å, b = 16.091(3) Å, c = 26.432(5) Å, β = 91.03(3)°, V = 5909.7 ų, Z = 4, Rgt(F) = 0.050, wRref(F²) = 0.127, T = 153 K
Stochastic multilevel methods for deep learning
No full textAs the use cases for neural networks become increasingly complex, modern neural networks must also grow deeper and more intricate to keep up. However, with increased depth and complexity, common training methods like stochastic gradient descent methods (SGD) become more computationally expensive.
Multilevel methods, traditionally used to solve differential equations through hierarchies of discretizations, offer the potential to reduce computational effort.
These methods operate across multiple levels of complexity. On lower levels, a coarse approximation of the solution can be obtained quickly and at low computational cost. These coarse approximations are then refined on higher levels to achieve a more accurate solution. By alternating between levels, multilevel methods can accelerate convergence while maintaining precision.
This thesis introduces a multilevel stochastic gradient descent algorithm (MLSGD) that combines both concepts aiming to accelerate neural network training through multilevel techniques. The core contribution of this thesis is the development and analysis of MLSGD. As in traditional multilevel methods, prolongation and restriction operators enable transitions between levels. To ensure first-order coherence, a gradient correction is added to the objective function as well as additional conditions including step size regularization and an angle condition.
We analyze the convergence properties of the method under the assumption of fixed step sizes. Additionally, we investigate the influence of stochastic directions in the gradient correction as a replacement for full gradients as well as the effect of variance reduction in both cases.
Finally, we evaluate the practical performance of the method as well as the effect of stochastic gradient correction and variance reduction. To this end, MLSGD is applied to the image classification dataset CIFAR-10. We explore two different approaches to constructing a hierarchy, based either on network depth or image resolution. For both approaches, we construct suitable prolongation and restriction operators
Implementation of quantum and classical discrete fractional Fourier transforms
Get PDFFourier transforms, integer and fractional, are ubiquitous mathematical tools in basic and applied science. Certainly, since the ordinary Fourier transform is merely a particular case of a continuous set of fractional Fourier domains, every property and application of the ordinary Fourier transform becomes a special case of the fractional Fourier transform. Despite the great practical importance of the discrete Fourier transform, implementation of fractional orders of the corresponding discrete operation has been elusive. Here we report classical and quantum optical realizations of the discrete fractional Fourier transform. In the context of classical optics, we implement discrete fractional Fourier transforms of exemplary wave functions and experimentally demonstrate the shift theorem. Moreover, we apply this approach in the quantum realm to Fourier transform separable and path-entangled biphoton wave functions. The proposed approach is versatile and could find applications in various fields where Fourier transforms are essential tools
Einflussfaktoren auf die Bondemissionsprämie im Bereich der alternativen Energien: Eine Note
Get PDFDie folgende Analyse präsentiert aktuelle empirische Evidenz, die es erlaubt, eine bessere Vorstellung der von den potentiellen Investoren geforderten Risikoprämie zum Emissionszeitpunkt zu erhalten. Untersucht werden zwischen 2008 und 2012 in Deutschland emittierte Anleihen im Bereich der alternativen Energien. Neben dem schon hinreichend bekannten Einfluss der Risikoklassifizierung durch ein Rating zeigen sich weitere bedeutende Variablen. Mit zunehmenden Emissionsvolumina und höherer Stückelung geht ebenfalls der Zinsaufschlag zurück. Offensichtlich konnten besonders große an institutionelle Investoren gerichtete Branchenvertreter in der Vergangenheit von günstigen Anleiheemissionen profitieren. Ebenfalls für eine niedrigere Emissionsprämie sorgen eine positive Konjunkturlage und ein hohes risikoloses Zinsniveau. Dies macht deutlich, dass das derzeitig in Deutschland herrschende Niedrigzinsniveau gepaart mit mäßigem Wachstum die kostengünstige Aufnahme von Fremdkapital über den Kapitalmarkt deutlich erschwert. Eine Durchsetzung der Maßnahmen der durch BMU und BMWi angekündigten „Strompreisbremse" würde die Situation für Unternehmen noch verschärfen
The Alternative Three-Factor Model: Evidence from the German Stock Market
Get PDFThis article applies the alternative three-factor model introduced by Chen / Novy-Marx / Zhang (2010) to the German stock market for the sample period of 2004 through 2015. We construct two new factors INV („investment") and ROA („return on assets") for companies listed on the highest segment of the Frankfurt Stock Exchange, and examine whether they can explain various stock market anomalies using linear time series regressions. Our results reveal that the theoretical assumptions of the model are valid for the German stock market. Firms with higher investments generally exhibit lower returns, while more profitable firms exhibit higher returns. However, we find that the alternative three-factor model does not explain capital market anomalies in the German market better than the factors of the traditional Fama / French (1993) three-factor model
The Importance of Metal–Support Interactions for CO₂ Hydrogenation: An Operando Near Ambient Pressure X-ray Photoelectron Spectroscopy Study on Gold-Loaded In₂O₃ and CeO₂ Catalysts
Get PDFMetal–support interactions, which are essential for the design of supported metal catalysts, used, e.g., for CO₂ activation, are still only partially understood. In this study of gold-loaded In₂O₃ and CeO₂ catalysts during CO₂ hydrogenation using near-ambient pressure X-ray photoelectron spectroscopy, supported by near edge X-ray absorption fine structure, we demonstrate that the role of the noble metal strongly depends upon the choice of the support material. Temperature-dependent analyses of X-ray photoelectron spectra under reaction conditions reveal that gold is reduced on CeO₂, enabling direct H₂ activation, but oxidized on In₂O₃, leading to decreased activity of Au/In₂O₃ compared to bare In₂O₃. At elevated temperatures, the catalytic activity of the In₂O₃ catalysts strongly increases as a result of facilitated CO₂ and (In₂O₃-based) H₂ activation, while the catalytic activity of Au/CeO₂ is limited by reoxidation by CO₂. Our results underline the importance of operando studies for understanding metal–support interactions to enable a rational support selection in the future
Where Are We? Security, Misuse, and Manipulation of Location-based Systems
Get PDFLocation-based systems allow us to perform many tasks: we can check the weather wherever we are, navigate unfamiliar environments, and track our running workouts as well as valuables such as laptops or luggage. However, all of these examples require access to highly sensitive and private location data. Misuse of this data allows adversaries to identify individuals based on a few locations, track them digitally without consent, or manipulate location data to bypass location-based access policies, such as those implemented in cars or smart door locks.
In this thesis, we contribute to three emerging areas of location-based systems: (1) Security and misuse of Bluetooth offline-finding networks; (2) Security, reliability, and accuracy of localization with UWB; and (3) Secure location sharing via satellite. Our methodology is diverse. We reverse-engineer implementations to investigate their privacy and security protections, open up proprietary systems and provide open-source implementations. Furthermore, we develop tools to protect users from unwanted location tracking, research novel attack vectors and protections, and conduct user studies to understand the needs of victims of unwanted tracking.
To summarize our main research results: (1) We analyze Apple’s Find My network, uncover vulnerabilities that expose individuals’ location data, and demonstrate that the network can be misused for tracking and stalking. Based on our results, we open up the proprietary network and publish a framework that allows any programmable Bluetooth device to be integrated into the Find My network, making it locatable. To protect users from unwanted tracking attacks, we develop and publish the AirGuard app, which identifies malicious trackers following a person and alerts them. We conduct a survey with over 5,000 participants and find that 44% of stalking victims have suffered from unwanted location tracking. With the study results and AirGuard user data, we identify patterns and methods used by stalkers, and where trackers are hidden. (2) We measure the accuracy and reliability of Ultra-Wide Band (UWB) distance measurements in consumer smartphones and find that, while results are accurate in most cases, all devices measured a small number of large outliers. Furthermore, we analyze the security of UWB against distance manipulation attacks and develop a novel physical-layer attack that can reduce the measured distance by several meters. (3) Encryption and privacy protection mechanisms are often neglected in satellite communication. We evaluate the iPhone’s location-sharing via satellite feature and find robust security mechanisms with multi-layer encryption in place. Nevertheless, we identify bypasses around Apple’s restrictions and manage to use satellite services from restricted areas and send text messages encoded in location updates. We propose countermeasures for all vulnerabilities we find.
In addition to academic publications, our work has demonstrated real-world impact with the AirGuard apps for Android and iOS. They have been installed over 1.3million times in the last three and a half years. Several victims of unwanted tracking informed us that they had found trackers using our application, preventing further tracking. By pushing for better detection mechanisms, we influenced major companies: Apple and Google have proposed standardized tracking detection for all Bluetooth trackers
The physics of defect chemistry and the chemistry of defect physics
Get PDFDefect chemistry is the classical approach to evaluate point-defect concentrations in solids depending on the chemical activity of the (n−1) of n constituents by evaluating the mass action laws of a number of defect reactions conserving species, lattice sites, and charge. In an alternative approach, formation energies of individual defects can be calculated to determine the dependence on the Fermi level and on the chemical potentials of the reservoirs. This contribution provides the quantitative relationship between the two approaches, offering the opportunity to compare calculated defect formation energies with experimentally determined quantities. As an example, the application of the two approaches to the comparison of electronic and ionic compensation of doping and the influence of the band edge energies on it is given. This example demonstrates that the Gibbs energy of reduction and oxidation are essentially aligning the energy axis of ionic defects relative to that of electronic defects. In conjunction with the dependence of the valence band maximum and conduction band minimum energies on material composition, this offers the opportunity to tune the preference for electronic, ionic or mixed compensation of doping by two independent quantities
A Concept for Shared Control of Unmanned Aircraft Systems
Get PDFWith the rise of agile air mobility, combined with low operating costs and advanced autonomous technologies, unmanned aircraft systems (UAS) are increasingly being integrated into various industries, including transportation, agriculture, and media. Typically, UAS guidance is managed either by autonomous systems or pilots. Autonomous systems offer advantages like stability and quick responsiveness but may lack effective emergency handling. Conversely, pilots excel in environmental perception and adaptive learning but are susceptible to fatigue.
Inspired by this contradiction, the concept of shared control is introduced to harmonize these two control modes, leveraging their strengths while mitigating weaknesses. Following an extensive review of relevant literature, a rigorous definition of shared control for UAS operation is established, distinguishing it from assistant and filtering control methods.
Two real-world scenarios are selected to assess the feasibility of shared control. In the first scenario, collision avoidance, a comprehensive shared control system is developed. The pilot's control authority is dynamically allocated based on collision risk, seamlessly integrating commands from both the pilot and autonomous systems. Field experiments validate the system's effectiveness, while Monte Carlo simulations further demonstrate its ability to increase the probability of avoidance, revealing potential additional collision risks. Additionally, user experiments are conducted to evaluate user satisfaction with the system.
In the second scenario, aerial cinematography, a novel shared control method is developed based on optimal control methodologies. This approach incorporates a human input model and utilizes model predictive control techniques to build the shared control system. Through simulated flights, all intended functionalities are effectively realized. After completing the simulated aerial cinematography tasks, participants provide objective evaluations that align with anticipated satisfaction levels.
The next step involves exploring the applicability of shared control across various scenarios. Future shared control systems should also be optimized for seamless integration with autonomous systems that possess learning capabilities. Additionally, mitigating the potential risks associated with implementing shared control remains critical
A Holistic Framework for Developing Expert Systems to Improve Energy Efficiency in Manufacturing
Get PDFAmid growing environmental and societal concerns about energy use, companies face increasing pressure to adopt sustainable manufacturing practices. The European Union’s guiding principles, aimed in part at achieving climate neutrality and fostering green growth, underscore the need for systematic, data-driven approaches to energy efficiency. This involves the measurement, monitoring, and analysis of energy data. However, identifying efficiency potentials often relies on expert knowledge, which is becoming increasingly scarce due to skilled labor shortages. Expert systems offer a solution by consolidating and analyzing data to automatically identify energy-saving opportunities. These systems leverage stored expertise, applying it to measurement data to generate actionable insights, while their explicit knowledge representation and transparent reasoning facilitate knowledge transfer. Despite their potential, most expert systems are developed intuitively and tailored to specific applications, limiting their broader adoption. To address this, we propose a holistic framework for systematic expert system development, supported by defined personas and an expert system shell serving as a software template. The framework is demonstrated and evaluated through its application in a metalworking process chain