16580 research outputs found
Sort by
Determining the innermost structure of quasars by microlensing: Measuring, simulating and interpreting light curves of multiple quasars
No full textThis thesis aims at measuring quasar microlensing light curves and applying them to constrain the structure of quasar accretion disks. Thus, data was taken in two photometric filters at the Las Cumbres Observatory, using their global network of 1m telescopes since 2014. In the first part, applying difference imaging analysis together with point spread function photometry aided with Gaia data, we measure the light curves of the multiple images of eight gravitationally lensed quasars in the R and V band, covering almost ten years with in total 1872 epochs. For each quasar, we determine difference curves of the time delay corrected light curves. This removes the intrinsic quasar brightness variations present in all images with only uncorrelated microlensing variability of the individual images remaining. We find these additional variations, attributed to the source size depended microlensing of the individual images by compact objects in the lens galaxy, throughout our whole data set.
For the second part of this thesis, we focus on the prominent microlensing signal in image B of the quadruple quasar HE0435-1223, revealed through our difference curves. The variations appear to be chromatic, i.e. depend on the filter, with higher amplitude fluctuations in the V band. This is expected, since the hotter central region of the accretion disk experiences more microlensing variation due to its smaller size. To quantify this observation, by means of microlensing simulations, we are able to infer that the accretion disk of HE0435-1223 is indeed larger in radius by factors of 1.24^{+0.08}_{−0.20}, 1.42^{+0.11}_{−0.22} and 1.43^{+0.10}_{−0.23} in the R with respect to the V band, depending on the disk model, in agreement with the expectation from thin accretion disk theory, though with a tendency towards a shallower temperature profile. Additionally, we find disk half-light radii of 0.7 to 1.0 Einstein radii, corresponding to average inclined disk scale radii of around log ⟨R2500/cm⟩ ≃ 16.4^{+0.5}_{−0.7} at 2500Å in the quasar rest-frame
Charge and Energy Transport in Disordered Semiconductors
No full textUnderstanding charge transport in disordered materials like organic semiconductors (OSCs) and quantum dot (QD) solids is crucial for optimizing device applications. Although governed by the common physics of hopping transport, the origins of disorder differ significantly between these systems. This thesis explores the consequences of this shared physics under both equilibrium and non- equilibrium conditions. At equilibrium, distinct transport mechanisms are demonstrated: In ZnO QDs, temperature-dependent conductivity is dictated by the size-dependent charging energy distribution, while in highly doped OSCs, extreme density of state filling induces a hard-Coulomb gap, leading to the novel phenomenon of Seebeck coefficient inversion. Under high electric fields, both systems exhibit non-equilibrium "hot" carrier effects, which are described using an effective electronic temperature framework. A physically grounded heat balance model is employed to extract an effective localization length (α_eff) from field-dependent conductivity, proving the potential of α_eff as a sensitive probe of morphology in both material classes. The central contribution is the direct experimental validation of the effective temperature concept by combining direct Seebeck-based thermometry with conventional conductivity scaling in ZnO QD solids. It is then proven that the effective temperature corresponds to the real, physical temperature of the non-equilibrium carrier distribution. This confirmation establishes the effective temperature framework as a reliable tool for studying non-equilibrium phenomena in disordered systems. Finally, the research connects fundamental understanding to material processing, demonstrating how techniques like dip-coating can modify local microstructure to enhance the thermoelectric performance of OSCs
Das Absolute und die Frage: Wozu? - Impuls
No full textDer bloßen Nennung des Begriffs „das Absolute“ folgen oft mehrere Repliken. Daher scheint es mir geboten, sich wenigstens kurz direkt mit ihnen auseinanderzusetzen. Drei Fragen werde ich thematisieren: (1) Leben wir in einem postmetaphysischen Zeitalter? (2) Ist das Absolute ein religiöser Begriff? (3) Brauchen wir das Absolute? Insgesamt lässt sich wie folgt antworten: Die philosophische Beschäftigung mit dem Absoluten weist eine enorme Bedeutung auf und ist alles andere als obsolet oder sinnlos
Resource-Efficient and Robust Inference of Deep and Bayesian Neural Networks on Embedded and Analog Computing Platforms
No full textWhile modern machine learning has transformed numerous application domains, its growing computational demands increasingly constrain scalability and efficiency, particularly on embedded and resource-constrained platforms. In practical deployments, neural networks must not only operate efficiently but also provide reliable predictions when faced with distributional changes or previously unseen data. Bayesian neural networks offer a principled framework for quantifying uncertainty, but their higher computational demands further compound these challenges.
This work advances resource-efficient and robust inference for both conventional and Bayesian neural networks through the joint pursuit of algorithmic and hardware efficiency. The former reduces computational cost through model compression and approximate Bayesian inference, while the latter optimizes mapping to digital accelerators and explores novel analog hardware platforms, bridging algorithmic optimization and physical realization.
The first contribution introduces the Galen framework, which performs automatic, layer-specific compression guided by sensitivity analysis and hardware-in-the-loop feedback, jointly optimizing quantization and pruning to balance accuracy and efficiency on embedded devices. As analog accelerators offer additional efficiency gains at the cost of noise, their modeling exposes device imperfections, while a layer-wise analysis reveals how networks learn to tolerate such effects during training. This work extends noisy training to nonstationary conditions, thereby enhancing robustness and stability in analog hardware.
A complementary line of work advances probabilistic inference. Building on insights into Bayesian-neural-network design and training, this work develops analytic and ensemble-based approximations that replace costly sampling, integrates them into a compiler stack, and optimizes them for probabilistic inference on embedded hardware. Finally, probabilistic photonic computing introduces a novel paradigm in which controlled analog noise serves as an intrinsic entropy source, enabling ultrafast and energy-efficient probabilistic inference directly in hardware.
Together, these studies demonstrate how efficiency and reliability can be advanced jointly through the co-design of algorithms, compilers, and hardware, laying the foundation for the next generation of trustworthy and energy-efficient machine-learning systems
How to connect the dots very fast? High-Performance Heterogeneous Particle Track Reconstruction for the ATLAS Phase-II High Level Trigger
No full textThe ATLAS experiment is a key project in high-energy particle physics exploring particle collisions at unprecedented energies to recreate early-universe conditions and probe phenomena that occur at extreme scales. The upcoming high luminosity upgrade of the LHC will significantly increase the collision rate and energy; the higher data volume and collision complexity necessitate a major upgrade of the ATLAS detector to consist of a more granular tracking system. Due to much higher data rate and processing complexity, it is crucial to optimise the collision selection system (trigger system) and its most computationally expensive component - the track reconstruction algorithms. This work explores advanced optimisation methods, including graphics card acceleration and machine learning, to enhance computational efficiency and effectively manage the increased data throughput and complexity of the upgraded detector.
The first considered approach optimises the track reconstruction algorithm used in the ATLAS trigger system. By employing the track seeding on the graphic card accelerator and adjusting the track seed selection criteria, the final performance was improved by 95%, achieving an average processing time per event of 1.16 s. The performance was evaluated on different graphics cards, considering their limitations, with NVIDIA RTX 5000 Ada achieving the best results due to its exceptionally high number of processing cores.
The second part of this work focuses on the application of machine learning techniques to particle track reconstruction. A novel Interaction Graph Neural Network (IGNN) demonstrates
competitive reconstruction accuracy; however, it is known to be resource-consuming. To address these computational challenges, two optimisation strategies are proposed, aimed at reducing both memory consumption and inference time without compromising model performance.
The instantaneous memory footprint of the model was reduced by partial processing (substepping). Memory consumption can be decreased by approximately 30% without an increase in processing time. Further memory reductions are achievable by adjusting the size of partitions, enabling the deployment of the IGNN on memory-constrained GPUs and allowing parallel processing, depending on the available hardware resources.
The second discussed compression technique is structured pruning of IGNN, where by removing the least important groups of parameters, the model size is reduced. A selection of pruning techniques applied to Graph Neural Networks (GNN) was analysed to determine the most effective methodology for GNN compression. The final pruning configuration achieves up to 20% improvement in computational performance without compromising model accuracy. Furthermore, per-layer sensitivity was analysed and incorporated in the pruning strategy to guide layer-wise pruning aggressiveness, enabling further model size reduction by 20% while maintaining reconstruction accuracy. The pruning strategy was evaluated on the standard GNN benchmark models, demonstrating satisfying performance gains. The performance of IGNN was evaluated on different graphics cards, considering their limitations, with NVIDIA RTX A100 achieving the best results due to its highly efficient memory throughput
Reaping the Benefits of Water: A Transcultural History of Technology and Empire in Ming Yunnan
No full textThis dissertation examines the Ming Empire’s rule in Yunnan through the dual lenses of technology and transculturality, with water control serving as its central focus. It argues that shuili 水利—conventionally translated as “water conservancy”—functioned not as a neutral technical domain but as a historically specific technological culture: a configuration of material practices, technical skills, labor organization, cultural norms, and moral meanings forged in the irrigated landscapes of Jiangnan and subsequently projected as a universalizing template for imperial rule.
The study traces the history of this shuili during the eleventh-century Song reforms, matured through Southern Song land reclamation and waterwork construction, and traveled to Yunnan as part of early Ming’s colonization project. There, it encountered resilient indigenous hydrosocial orders centered on hillside watertanks, fragmented landholding, and ritual authority vested in dragons, Buddhist acharyas, and lineage networks. Rather than a simple imposition of “Chinese” techniques, Yunnan's hydraulic transformation emerged through negotiated encounters: imported Jiangnan norms reshaped local ecologies, while being constrained and reworked by environmental limits, cosmological geographies, and the uneven distribution of technical resources.
By the late sixteenth century, this colonial graft revealed structural limits. Provincial decision-making centralized even as hydrological knowledge remained diffuse and embodied in local practice. The resulting crisis exposed how imperial technological cultures are vulnerable to information asymmetries, maintenance failures, and the fragility of transmitted skill. Treating shuili as mobile, contested, and hybrid, the dissertation offers a transcultural history of technology that grounds imperial statecraft in the material, organizational, and symbolic work of governing water
Parameter Estimation and Prediction in Nonlinear Dynamics
No full textNonlinear dynamical systems are central to modeling complex phenomena across science and engineering, from biological networks to climate systems. A critical challenge in these
systems is the accurate estimation of model parameters and the reliable prediction of future states, especially under uncertainty and limited observability. This thesis addresses this challenge by developing a proper framework that integrates classical methods, modern optimization theory, and machine learning techniques for parameter estimation and prediction in nonlinear dynamical systems.
We begin by revisiting foundational approaches to parameter estimation in ordinary differential equation models, including least squares, maximum likelihood, and Bayesian inference. To bridge the gap between theory and practice, we explore advanced computational techniques such as multiple shooting, collocation methods, and robust estimation with the Huber loss function.
Numerical optimization plays a central role in our methodology. The thesis presents detailed analyses of unconstrained and constrained optimization algorithms, including Newtonbased methods, trust-region strategies, and sequential quadratic programming. These methods are then applied to system identification tasks, where we contrast classical strategies with data-driven machine learning approaches.
In the latter part of the thesis, we propose hybrid methods that combine traditional system identification with deep learning architectures such as feedforward neural networks
and neural differential equations. We introduce a machine learning-based framework for parameter estimation, supported by theoretical analysis and extensive numerical experiments
on benchmark systems including the Van der Pol oscillator, Lotka-Volterra dynamics, and the Lorenz attractor.
Finally, we develop a real-time dynamic state estimation framework based on moving horizon estimation using the qpOASES solver and a reformulated Huber penalty function.
This method enables robust, online estimation in noisy environments and is further extended with a neural ODE and multiple shooting-based architecture.
Overall, the results underscore the critical role of accurate parameter estimation in improving the reliability of nonlinear system predictions, with implications for diverse domains including physics, biology, engineering, and finance