Refine
Year of publication
- 2024 (4)
- 2023 (7)
- 2022 (11)
- 2021 (10)
- 2020 (37)
- 2019 (156)
- 2018 (195)
- 2017 (117)
- 2016 (73)
- 2015 (18)
- 2014 (11)
- 2013 (17)
- 2012 (25)
- 2011 (21)
- 2010 (6)
- 2009 (7)
- 2008 (12)
- 2007 (11)
- 2006 (27)
- 2005 (17)
- 2004 (11)
- 2003 (4)
- 2002 (3)
- 2001 (2)
- 2000 (4)
- 1999 (2)
- 1998 (10)
- 1997 (5)
- 1996 (13)
- 1995 (9)
- 1994 (15)
- 1993 (4)
- 1992 (3)
- 1991 (1)
Document Type
- Other (869) (remove)
Language
- English (638)
- German (217)
- Spanish (5)
- Italian (3)
- Multiple languages (2)
- Polish (2)
- French (1)
- Portuguese (1)
Keywords
- Arrayseismologie (5)
- array seismology (5)
- Dysphagie (4)
- E-Learning (4)
- Erdbeben (4)
- Judaism (4)
- Judentum (4)
- MOOC (4)
- Patholinguistik (4)
- Schluckstörung (4)
Institute
- Institut für Biochemie und Biologie (85)
- Hasso-Plattner-Institut für Digital Engineering GmbH (83)
- Institut für Physik und Astronomie (82)
- Institut für Geowissenschaften (72)
- Institut für Mathematik (46)
- Department Psychologie (44)
- Department Sport- und Gesundheitswissenschaften (44)
- Institut für Ernährungswissenschaft (31)
- Hasso-Plattner-Institut für Digital Engineering gGmbH (30)
- Institut für Chemie (29)
Introduction
(2018)
Detecting and recognizing text in natural scene images is a challenging, yet not completely solved task. In recent years several new systems that try to solve at least one of the two sub-tasks (text detection and text recognition) have been proposed. In this paper we present SEE, a step towards semi-supervised neural networks for scene text detection and recognition, that can be optimized end-to-end. Most existing works consist of multiple deep neural networks and several pre-processing steps. In contrast to this, we propose to use a single deep neural network, that learns to detect and recognize text from natural images, in a semi-supervised way. SEE is a network that integrates and jointly learns a spatial transformer network, which can learn to detect text regions in an image, and a text recognition network that takes the identified text regions and recognizes their textual content. We introduce the idea behind our novel approach and show its feasibility, by performing a range of experiments on standard benchmark datasets, where we achieve competitive results.
We propose a new temporal extension of the logic of Here-and-There (HT) and its equilibria obtained by combining it with dynamic logic over (linear) traces. Unlike previous temporal extensions of HT based on linear temporal logic, the dynamic logic features allow us to reason about the composition of actions. For instance, this can be used to exercise fine grained control when planning in robotics, as exemplified by GOLOG. In this paper, we lay the foundations of our approach, and refer to it as Linear Dynamic Equilibrium Logic, or simply DEL. We start by developing the formal framework of DEL and provide relevant characteristic results. Among them, we elaborate upon the relationships to traditional linear dynamic logic and previous temporal extensions of HT.
3D point cloud technology facilitates the automated and highly detailed digital acquisition of real-world environments such as assets, sites, cities, and countries; the acquired 3D point clouds represent an essential category of geodata used in a variety of geoinformation applications and systems. In this paper, we present a web-based system for the interactive and collaborative exploration and inspection of arbitrary large 3D point clouds. Our approach is based on standard WebGL on the client side and is able to render 3D point clouds with billions of points. It uses spatial data structures and level-of-detail representations to manage the 3D point cloud data and to deploy out-of-core and web-based rendering concepts. By providing functionality for both, thin-client and thick-client applications, the system scales for client devices that are vastly different in computing capabilities. Different 3D point-based rendering techniques and post-processing effects are provided to enable task-specific and data-specific filtering and highlighting, e.g., based on per-point surface categories or temporal information. A set of interaction techniques allows users to collaboratively work with the data, e.g., by measuring distances and areas, by annotating, or by selecting and extracting data subsets. Additional value is provided by the system's ability to display additional, context-providing geodata alongside 3D point clouds and to integrate task-specific processing and analysis operations. We have evaluated the presented techniques and the prototype system with different data sets from aerial, mobile, and terrestrial acquisition campaigns with up to 120 billion points to show their practicality and feasibility.
Business processes constantly generate, manipulate, and consume data that are managed by organizational databases. Despite being central to process modeling and execution, the link between processes and data is often handled by developers when the process is implemented, thus leaving the connection unexplored during the conceptual design. In this paper, we introduce, formalize, and evaluate a novel conceptual view that bridges the gap between process and data models, and show some kinds of interesting insights that can be derived from this novel proposal.
We present a prototype of an integrated reasoning environment for educational purposes. The presented tool is a fragment of a proof assistant and automated theorem prover. We describe the existing and planned functionality of the theorem prover and especially the functionality of the educational fragment. This currently supports working with terms of the untyped lambda calculus and addresses both undergraduate students and researchers. We show how the tool can be used to support the students' understanding of functional programming and discuss general problems related to the process of building theorem proving software that aims at supporting both research and education.
Declarative languages for knowledge representation and reasoning provide constructs to define preference relations over the set of possible interpretations, so that preferred models represent optimal solutions of the encoded problem. We introduce the notion of approximation for replacing preference relations with stronger preference relations, that is, relations comparing more pairs of interpretations. Our aim is to accelerate the computation of a non-empty subset of the optimal solutions by means of highly specialized algorithms. We implement our approach in Answer Set Programming (ASP), where problems involving quantitative and qualitative preference relations can be addressed by ASPRIN, implementing a generic optimization algorithm. Unlike this, chains of approximations allow us to reduce several preference relations to the preference relations associated with ASP’s native weak constraints and heuristic directives. In this way, ASPRIN can now take advantage of several highly optimized algorithms implemented by ASP solvers for computing optimal solutions
Manufacturing industries are undergoing a major paradigm shift towards more autonomy. Automated planning and scheduling then becomes a necessity. The Planning and Execution Competition for Logistics Robots in Simulation held at ICAPS is based on this scenario and provides an interesting testbed. However, the posed problem is challenging as also demonstrated by the somewhat weak results in 2017. The domain requires temporal reasoning and dealing with uncertainty. We propose a novel planning system based on Answer Set Programming and the Clingo solver to tackle these problems and incentivize robot cooperation. Our results show a significant performance improvement, both, in terms of lowering computational requirements and better game metrics.
DualPanto
(2018)
We present a new haptic device that enables blind users to continuously track the absolute position of moving objects in spatial virtual environments, as is the case in sports or shooter games. Users interact with DualPanto by operating the me handle with one hand and by holding on to the it handle with the other hand. Each handle is connected to a pantograph haptic input/output device. The key feature is that the two handles are spatially registered with respect to each other. When guiding their avatar through a virtual world using the me handle, spatial registration enables users to track moving objects by having the device guide the output hand. This allows blind players of a 1-on-1 soccer game to race for the ball or evade an opponent; it allows blind players of a shooter game to aim at an opponent and dodge shots. In our user study, blind participants reported very high enjoyment when using the device to play (6.5/7).
Scenograph
(2018)
When developing a real-walking virtual reality experience, designers generally create virtual locations to fit a specific tracking volume. Unfortunately, this prevents the resulting experience from running on a smaller or differently shaped tracking volume. To address this, we present a software system called Scenograph. The core of Scenograph is a tracking volume-independent representation of real-walking experiences. Scenograph instantiates the experience to a tracking volume of given size and shape by splitting the locations into smaller ones while maintaining narrative structure. In our user study, participants' ratings of realism decreased significantly when existing techniques were used to map a 25m2 experience to 9m2 and an L-shaped 8m2 tracking volume. In contrast, ratings did not differ when Scenograph was used to instantiate the experience.
TrussFormer
(2018)
We present TrussFormer, an integrated end-to-end system that allows users to 3D print large-scale kinetic structures, i.e., structures that involve motion and deal with dynamic forces. TrussFormer builds on TrussFab, from which it inherits the ability to create static large-scale truss structures from 3D printed connectors and PET bottles. TrussFormer adds movement to these structures by placing linear actuators into them: either manually, wrapped in reusable components called assets, or by demonstrating the intended movement. TrussFormer verifies that the resulting structure is mechanically sound and will withstand the dynamic forces resulting from the motion. To fabricate the design, TrussFormer generates the underlying hinge system that can be printed on standard desktop 3D printers. We demonstrate TrussFormer with several example objects, including a 6-legged walking robot and a 4m-tall animatronics dinosaur with 5 degrees of freedom.
Low back pain (LBP) is a leading cause of activity limitation. Objective assessment of the spinal motion plays a key role in diagnosis and treatment of LBP. We propose a method that facilitates clinical assessment of lower back motions by means of a wireless inertial sensor network. The sensor units are attached to the right and left side of the lumbar region, the pelvis and the thighs, respectively. Since magnetometers are known to be unreliable in indoor environments, we use only 3D accelerometer and 3D gyroscope readings. Compensation of integration drift in the horizontal plane is achieved by estimating the gyroscope biases from automatically detected initial rest phases. For the estimation of sensor orientations, both a smoothing algorithm and a filtering algorithm are presented. From these orientations, we determine three-dimensional joint angles between the thighs and the pelvis and between the pelvis and the lumbar region. We compare the orientations and joint angles to measurements of an optical motion tracking system that tracks each skin-mounted sensor by means of reflective markers. Eight subjects perform a neutral initial pose, then flexion/extension, lateral flexion, and rotation of the trunk. The root mean square deviation between inertial and optical angles is about one degree for angles in the frontal and sagittal plane and about two degrees for angles in the transverse plane (both values averaged over all trials). We choose five features that characterize the initial pose and the three motions. Interindividual differences of all features are found to be clearly larger than the observed measurement deviations. These results indicate that the proposed inertial sensor-based method is a promising tool for lower back motion assessment.
Beware of SMOMBIES
(2018)
Several research evaluated the user's style of walking for the verification of a claimed identity and showed high authentication accuracies in many settings. In this paper we present a system that successfully verifies a user's identity based on many real world smartphone placements and yet not regarded interactions while walking. Our contribution is the distinction of all considered activities into three distinct subsets and a specific one-class Support Vector Machine per subset. Using sensor data of 30 participants collected in a semi-supervised study approach, we prove that unsupervised verification is possible with very low false-acceptance and false-rejection rates. We furthermore show that these subsets can be distinguished with a high accuracy and demonstrate that this system can be deployed on off-the-shelf smartphones.
S-test results for the USGS and RELM forecasts. The differences between the simulated log-likelihoods and the observed log-likelihood are labelled on the horizontal axes, with scaling adjustments for the 40year.retro experiment. The horizontal lines represent the confidence intervals, within the 0.05 significance level, for each forecast and experiment. If this range contains a log-likelihood difference of zero, the forecasted log-likelihoods are consistent with the observed, and the forecast passes the S-test (denoted by thin lines). If the minimum difference within this range does not contain zero, the forecast fails the S-test for that particular experiment, denoted by thick lines. Colours distinguish between experiments (see Table 2 for explanation of experiment durations). Due to anomalously large likelihood differences, S-test results for Wiemer-Schorlemmer.ALM during the 10year.retro and 40year.retro experiments are not displayed. The range of log-likelihoods for the Holliday-et-al.PI forecast is lower than for the other forecasts due to relatively homogeneous forecasted seismicity rates and use of a small fraction of the RELM testing region.
In an effort to explain the formation of a narrow third radiation belt at ultra-relativistic energies detected during a solar storm in September 20121, Mann et al.2 present simulations from which they conclude it arises from a process of outward radial diffusion alone, without the need for additional loss processes from higher frequency waves. The comparison of observations with the model in Figs 2 and 3 of their Article clearly shows that even with strong radial diffusion rates, the model predicts a third belt near L* = 3 that is twice as wide as observed and approximately an order of magnitude more intense. We therefore disagree with their interpretation that “the agreement between the absolute fluxes from the model and those observed by REPT [the Relativistic Electron Proton Telescope] shown on Figs 2 and 3 is excellent.”
Previous studies3 have shown that outward radial diffusion plays a very important role in the dynamics of the outer belt and is capable of explaining rapid reductions in the electron flux. It has also been shown that it can produce remnant belts (Fig. 2 of a long-term simulation study4). However, radial diffusion alone cannot explain the formation of the narrow third belt at multi-MeV during September 2012. An additional loss mechanism is required.
Higher radial diffusion rates cannot improve the comparison of model presented by Mann et al. with observations. A further increase in the radial diffusion rates (reported in Fig. 4 of the Supplementary Information of ref. 2) results in the overestimation of the outer belt fluxes by up to three orders of magnitude at energy of 3.4 MeV.
Observations at 2 MeV, where belts show only a two-zone structure, were not presented by Mann et al. Moreover, simulations of electrons with energies below 2 MeV with the same diffusion rates and boundary conditions used by the authors would probably produce very strong depletions down to L = 3–3.5, where L is radial distance from the centre of the Earth to the given field line in the equatorial plane. Observations do not show a non-adiabatic loss below L ∼ 4.5 for 2 MeV. Such different dynamics between 2 MeV and above 4 MeV at around L = 3.5 are another indication that particles are scattered by electromagnetic ion cyclotron (EMIC) waves that affect only energies above a certain threshold.
Observations of the phase space density (PSD) provide additional evidence for the local loss of electrons. Around L* = 3.5–4 PSD shows significant decrease by an order of magnitude starting in the afternoon of 3 September (Fig. 1a), while PSD above L* = 4 is increasing. The minimum in PSD between L* = 3.5–4 continues to decrease until 4 September. This evolution demonstrates that the loss is not produced by outward diffusion. Radial diffusion cannot produce deepening minima, as it works to smooth gradients. Just as growing peaks in PSD show the presence of localized acceleration5, deepening minima show the presence of localized loss.
Figure 1: Time evolution of radiation profiles in electron PSD at relativistic and ultra-relativistic energies.
figure 1
a, Similar to Supplementary Fig. 3 of ref. 2, but using TS07D model10 and for μ = 2,500 MeV G−1, K = 0.05 RE G0.5 (where RE is the radius of the Earth). b, Similar to Supplementary Fig. 3 of ref. 2, but using TS07D model and for μ = 700 MeV G−1, corresponding to MeV energies in the heart of the belt. Minimum in PSD in the heart of the multi-MeV electron radiation belt between 3.5 and 4 RE deepening between the afternoon of 3 September and 5 September clearly show that the narrow remnant belt at multi-MeV below 3.5 RE is produced by the local loss.
Full size image
The minimum in the outer boundary is reached on the evening of 2 September. After that, the outer boundary moves up, while the minimum decreases by approximately an order of magnitude, clearly showing that this main decrease cannot be explained by outward diffusion, and requires additional loss processes. The analysis of profiles of PSD is a standard tool used, for example, in the study about electron acceleration5 and routinely used by the entire Van Allen Probes team. In the Supplementary Information, we show that this analysis is validated by using different magnetic field models. The Supplementary Information also shows that measurements are above background noise.
Deepening minima at multi-MeV during the times when the boundary flux increases are clearly seen in Fig. 1a. They show that there must be localized loss, as radial diffusion cannot produce a minimum that becomes lower with time. At lower energies of 1–2 MeV, which corresponds to lower values of the first adiabatic invariant μ (Fig. 1b), the profiles are monotonic between L* = 3–3.5, consistent with the absence of scattering by EMIC waves that affect only electrons above a certain energy threshold6,7,8,9.
In summary, the results of the modelling and observations presented by Mann et al. do not lend support to the claim of explaining the dynamics of the ultra-relativistic third Van Allen radiation belt in terms of an outward radial diffusion process alone. While the outward radial diffusion driven by the loss to the magnetopause2 is certainly operating during this storm, there is compelling observational and modelling2,6 evidence that shows that very efficient localized electron loss operates during this storm at multi-MeV energies, consistent with localized loss produced by EMIC waves.