4242 research outputs found
Sort by
Principles of Neural Network Architecture Design - Invertibility and Domain Knowledge
Get PDFNeural networks architectures allow a tremendous variety of design choices. In this work, we study two principles underlying these architectures: First, the design and application of invertible neural networks (INNs). Second, the incorporation of domain knowledge into neural network architectures. After introducing the mathematical foundations of deep learning, we address the invertibility of standard feedforward neural networks from a mathematical perspective. These results serve as a motivation for our proposed invertible residual networks (i-ResNets). This architecture class is then studied in two scenarios: First, we propose ways to use i-ResNets as a normalizing flow and demonstrate the applicability for high-dimensional generative modeling. Second, we study the excessive invariance of common deep image classifiers and discuss consequences for adversarial robustness. We finish with a study of convolutional neural networks for tumor classification based on imaging mass spectrometry (IMS) data. For this application, we propose an adapted architecture guided by our knowledge of the domain of IMS data and show its superior performance on two challenging tumor classification datasets
Bringing a Humanoid Robot Closer to Human Versatility : Hard Realtime Software Architecture and Deep Learning Based Tactile Sensing
Get PDFFor centuries, it has been a vision of man to create humanoid robots, i.e., machines that not only resemble the shape of the human body, but have similar capabilities, especially in dextrously manipulating their environment. But only in recent years it has been possible to build actual humanoid robots with many degrees of freedom (DOF) and equipped with torque controlled joints, which are a prerequisite for sensitively acting in the world. In this thesis, we extend DLR's advanced mobile torque controlled humanoid robot Agile Justin into two important directions to get closer to human versatility. First, we enable Agile Justin, which was originally built as a research platform for dextrous mobile manipulation, to also be able to execute complex dynamic manipulation tasks. We demonstrate this with the challenging task of catching up to two simultaneously thrown balls with its hands. Second, we equip Agile Justin with highly developed and deep learning based tactile sensing capabilities that are critical for dextrous fine manipulation. We demonstrate its tactile capabilities with the delicate task of identifying an objects material simply by gently sweeping with a fingertip over its surface. Key for the realization of complex dynamic manipulation tasks is a software framework that allows for a component based system architecture to cope with the complexity and parallel and distributed computational demands of deep sensor-perception-planning-action loops -- but under tight timing constraints. This thesis presents the communication layer of our aRDx (agile robot development -- next generation) software framework that provides hard realtime determinism and optimal transport of data packets with zero-copy for intra- and inter-process and copy-once for distributed communication. In the implementation of the challenging ball catching application on Agile Justin, we take full advantage of aRDx's performance and advanced features like channel synchronization. Besides developing the challenging visual ball tracking using only onboard sensing while everything is moving and the automatic and self-contained calibration procedure to provide the necessary precision, the major contribution is the unified generation of the reaching motion for the arms. The catch point selection, motion planning and the joint interpolation steps are subsumed in one nonlinear constrained optimization problem which is solved in realtime and allows for the realization of different catch behaviors. For the highly sensitive task of tactile material classification with a flexible pressure-sensitive skin on Agile Justin's fingertip, we present our deep convolutional network architecture TactNet-II. The input is the raw 16000 dimensional complex and noisy spatio-temporal tactile signal generated when sweeping over an object's surface. For comparison, we perform a thorough human performance experiment with 15 subjects which shows that Agile Justin reaches superhuman performance in the high-level material classification task (What material id?), as well as in the low-level material differentiation task (Are two materials the same?). To increase the sample efficiency of TactNet-II, we adapt state of the art deep end-to-end transfer learning to tactile material classification leading to an up to 15 fold reduction in the number of training samples needed. The presented methods led to six publication awards and award finalists and international media coverage but also worked robustly at many trade fairs and lab demos
Theoretical investigations of wide-bandgap semiconductor nanowires for optoelectronic applications
Get PDFImproving existing optoelectronic devices is a crucial step in satisfying humanity's increasing demand for electricity. This work explores different ways to achieve this goal. First density functional theory (DFT) calculations are performed on functionalized ZnO and GaN surface structures to investigate possible changes to their structural, electronic, and optical properties due to the attached functional groups. For both materials, attaching thiol groups leads to intra-gap states, which are found to be optically active for ZnO. Aiming at bigger GaN model sizes in future works compared to standard DFT approaches, a DFTB model was developed for GaN surface nanostructures. The interatomic interaction parameters were validated against standard DFT, achieving acceptable performances on bulk Ga, bulk GaN, and surface GaN systems. Another possible route to modify the electronic properties of semiconductor nanostructures is doping. ZnO bulk was doped with cobalt atoms to model different intrinsic defect complexes. Many-body GW calculations were employed to investigate their electronic structures. One defect complex is identified to be responsible for the experimentally observed photoluminescence. Due to the continuing decrease in size of electronic devices, the standard gate oxide SiO2 needs to be replaced, since today's required film thicknesses expose a crucial weakness of SiO2, a high tunneling leakage current. Possible candidates to be used as a replacement are hafnium silicate nanostructures, that avoid the described weakness. In a first step a density functional-based tight binding (DFTB) model for HfO2 was developed and validated against standard DFT calculations, achieving a very good performance for Hf bulk and HfO2 bulk. The obtained parameters were then used in a MD study on amorphous HfO2 systems to discuss their structural and electronic properties. In a second step this model was extended by silicon and applied to amorphous hafnium silicate structures to evaluate the influence of different Hf:Si ratios
Zugehörigkeitskonstruktionen in Domänen des Sprechens und Zeigens in einer teilintegrativen Willkommensklasse einer inklusiven Oberschule in X-Stadt. Ein reflexiver Aufsatz über ein empirisches Forschungsprojekt
Get PDFIm Rahmen dieses Artikels sollen die wesentlichen empirischen Befunde meiner Masterarbeit Zugehörigkeitskonstruktionen in Domänen des Sprechens und Zeigens in einer teilintegrativen Willkommensklasse einer inklusiven Oberschule in X-Stadt (Morthorst 2018, unveröffentlichte Masterarbeit an der Universität Bremen) sowie die theoretische Rahmung, Erhebungs- und Auswertungsmethoden und die Forschungsmethodologie dargestellt werden. Darüber hinaus erfolgt eine Reflexion der wesentlichen Schwierigkeiten, die mir im Rahmen der Konzeptualisierung der Forschungsarbeit begegnet sind. Es dreht sich dabei im Kern um die Schwierigkeit des Entdeckens bzw. Abstrahierens einer Kernkategorie und dem Treffen gegenstandstheoretischer Aussagen mit einer abduktiven Forschungshaltung. Des Weiteren werden vor dem Hintergrund der Kontextgebundenheit sozialer Interaktion, Fragen der Textorganisierung und des Suchens des Roten Fadens in der Ergebnispräsentation aufgeworfen. Daraufhin soll die Frage der Kompatibilität subjektivationstheoretischer Forschung in Verbindung mit der reflexiven Grounded Theory-Methodologie erörtert werden
Rapid Digital Architecture Design of Computationally Complex Algorithms
Get PDFTraditional digital design techniques hardly keep up with the rising abundance of programmable circuitry found on recent Field-Programmable Gate Arrays. Therefore, the novel Rapid Data Type-Agnostic Digital Design Methodology (RDAM) elevates the design perspective of digital design engineers away from the register-transfer level to the algorithmic level. It is founded on the capabilities of High-Level Synthesis tools. By consequently working with data type-agnostic source codes, the RDAM brings significant simplifications to the fixed-point conversion of algorithms and the design of complex-valued architectures. Signal processing applications from the field of Compressed Sensing illustrate the efficacy of the RDAM in the context of multi-user wireless communications. For instance, a complex-valued digital architecture of Orthogonal Matching Pursuit with rank-1 updating has successfully been implemented and tested
Quantum Frames and Uncertainty Principles arising from Symplectomorphisms
Get PDFIn this thesis, a new generalized signal transform along with a new uncertainty principle is elaborated. Starting from a coordinate system, associated to a specific symplectomorphism on phase space, the coordinates are used to define curvilinear flows along which the phase space picture of a prototype function is translated. As the uncertainty principle restricts the amount to which the phase space picture of functions can be concentrated, with each such function is associated a phase space cell. Using the phase space translates of these cells, the notion of a quantum frame is defined, by means of which a reservoir of interesting functions may be decomposed. To define optimal phase space cells, two complementing uncertainty principles, associated with coordinate systems in phase space, are introduced, one of which measures the deviation from the chosen frame, while the other optimizes with respect to the canonically conjugate coordinates and leads to more concentrated waveforms
Simulationsbasierte Entwicklung eines Kohlenstoff-Nanopartikel-basierten Heizsystems zur energieeffizienten Herstellung von kohlenstofffaserverstärkten Kunststoffbauteilen mit konturnah beheizten Formwerkzeugen
Get PDFThis work deals with the simulation-based development and implementation of an energy-efficient variant for the production and curing of fiber-reinforced plastics using a novel nanoparticle-based 3D tool tempering system. The current production process of CFRP components in the autoclave was analyzed with temperature measurements and flow simulations and various possibilities are presented to increase the efficiency by a conformal heating of the mold and reducing thermal masses respectively the overall heat capacity. Functional heated tool demonstrators, including electronic control technology, were designed and produced, as well as a process chain for the automated production or conversion of molds is shown. CFRP components were cured as a function test of the tool and the new process options were analyzed. In addition to the experimental heating system development, various process simulations on micro and macro scale were carried out. This work presents methods to support the different process steps from autoclave curing or curing with heated tools (composite curing with the finite element method), the definition of design parameters of the carbon nanotube reinforced heating system (electrical conductivity of nanocomposites) and the automated application onto the tool surface. The energy efficiency has been calculated for various processes (inside and out-of-autoclave) and it is shown that the energy consumption can be drastically reduced by the use of conformal heated tools. Finally, the developed 3D tool tempering system and the automated production were tested for a broader implementation and a profitability analysis was prepared
Schmierstofffreies Rundkneten / Trockenrundkneten
Get PDFIn this dissertation, the free-form forging process rotary swaging is investigated. Usually the established process is carried out under the use of a large amount of lubricants based on mineral oil. The lubricant fulfills important tasks for the process. The aim is the development of rotary swaging towards a dry process layout. To enable a dry process, it is necessary to substitute the functions of the lubricant; cooling of work piece and die, washing out of wear particles from the forming zone and process lubrication in general. The hypothesis of dry rotary swaging is to enable a dry process by means of coating and structuring of the dies. The structuring includes several size scales and the coating is constructed as a multi-layer system. This creates an optimized tribological condition in the different zones of the forging die. Finally, it is shown that such a tribologically adapted functionalized die enables dry rotary swaging of steel and aluminum workpieces
Control of Stereoselectivity in Heterogeneous Catalysis by Functionalization of Pt Nanoparticles with alpha-Amino Acids
No full textThe asymmetric hydrogenation of ß-keto esters over amino acid-functionalized Pt nanoparticles (NPs) was explored. The investigated catalysts are chemoselective and stereoselective due to specific ligand-reactant interactions between ligands and reactants. Based on structure-selectivity relationships, the previously for L-proline-functionalized Pt NPs suggested ligand-reactant interaction model could be validated, extended to other amino acid ligands, and further refined. Nonlinear effect analysis revealed that in most cases the stereoselectivity is determined mainly by a 1:1 ligand-reactant interaction. Adjusting the interactions between ligands and reactants allowed to improve the stereoselectivity up to an enantiomeric excess over 80%. Most amino acid ligands not only induce selectivity but also enhance the reaction rate. The results of a systematic kinetic study provided an important contribution to the understanding of the reaction mechanism
Paleoenvironmental evolution of on-shore ice-free areas around Maxwell Bay, King George Island, South Shetlands Islands
Get PDFUnderstanding the Holocene is particularly important for providing the context for recent ice sheet dynamics a i.e. understanding whether current ice sheet dynamics are unusual or part of Holocene natural variability (Bentley et al., 2014). Knowledge on the most recent millennia of Antarctic Ice Sheet history is vital for evaluating the response of the ice sheet to various forcing agents, such as sea-level rise, atmospheric and oceanographic temperature changes, and for constraining grounding-line retreat on Holocene to recent time scales (Bentley et al., 2014). The main objective of this thesis is to add new data to reconstruct the Holocene deglaciation history of King George Island, South Shetland Islands, northwest Antarctic Peninsula, by investigating morpho-sedimentary records of glacigenic and coastal landforms and associated sediments from the on-shore ice-free areas around Maxwell Bay (King George Island), namely Potter Peninsula and Fildes Peninsulas. In order to accomplish the thesis objectives, I used (i) cosmogenic exposure dating and radiocarbon dating for absolute chronological constraints; (ii) stratigraphy and sedimentology for relative chronological constraints and reconstruction of paleoenvironmental conditions; (iii) geomorphological mapping for spatial distribution of landsystems; (iv) and ground-penetrating radar (GPR) investigations for the study of internal sedimentary architecture of coastal landforms. Radiocarbon dating results yield new age constraints for the onset of deglaciation on Potter Peninsula, which occurred around at or before 7.8 ka cal BP instead of an earlier accepted age of 9.5 ka cal BP. I provide additional evidence for a short-lived glacier re-advance between 7.2 and 7.0 ka cal BP. This re-advance is likely linked to a glacier re-advance or still-stand documented on South Shetland Islands for that time period. Nevertheless, climatic conditions associated with this glacial re-advance remain unclear. In contrast, on Fildes Peninsula, exposure and radiocarbon dating indicate that glacial oscillations were minimal during the last 7 ka. I applied radiocarbon dating to remnants of mosses preserved in moraines. The moraines were formed close to the present glacier limit between 0.5 and 0.1 ka cal BP, during the last glacier re-advance in South Shetland Islands. This advance is linked to reductions in summer/annual insolation coupled with a shift to more intense Southern Hemisphere westerly winds in the Southern Ocean. Stronger, and possibly more poleward-shifted southern westerly winds produced more precipitation-laden storm fronts passing over the South Shetland Islands and thus, increased ice accumulation. The data also show that between 1.9 and 1.3 ka cal BP a climatic optimum was reached on Fildes and Potter Peninsula, which lasted until the last glacier readvance. GPR investigations and radiocarbon dating from a gravel spit system on Potter Peninsula document coastal progradation during the late phase of the last glacier re-advance, with a stable relative sea-level. Results also show an interruption of spit progradation that coincides with a proposed onset of accelerated isostatic rebound in reaction to glacier retreat subsequent to the last glacier re-advance. Spit growth resumed in the late 19th century after the rate of isostatic rebound decreased, and continues until today. The findings of this thesis support both, glacio-isostatic adjustment (GIA) models that show limited and those which show more pronounced ice-load changes on the South Shetland Islands during the late Holocene, suggesting that some GIA model parameters for the South Shetland Islands (e.g., lithospheric thickness, mantle viscosity) need to be better constrained. Furthermore, my findings have implications for regional paleoclimatic reconstructions and on ice sheet modeling for the Holocene of the northwest Antarctic Peninsula region