The Zuse Institute Berlin is a research institute for applied mathematics and computer science on the campus of Freie Universität Berlin in Dahlem.The ZIB was founded by law as a statutory establishment and as a non-university research institute of the State of Berlin in 1984. In close interdisciplinary cooperation with the Berlin universities and scientific institutions Zuse Institute implements research and development in the field of information technology with a particular focus on application-oriented algorithmic Mathematics and practical Computer science. ZIB also provides high-performance computer capacity as an accompanying service as part of the Network of high performance computers in Northern Germany ).Konrad Zuse, born in Berlin in 1910, is the namesake of the ZIB. Wikipedia.
Mukhopadhyay A.,Zuse Institute Berlin
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2017
This paper proposes a fully automatic supervised segmentation technique for segmenting the great vessel and blood pool of pediatric cardiac MRIs of children with Congenital Heart Defects (CHD). CHD affects the overall anatomy of heart, rendering model-based segmentation framework infeasible, unless a large dataset of annotated images is available. However, the cardiac anatomy still retains distinct appearance patterns, which has been exploited in this work. In particular, Total Variation (TV) is introduced for solving the 3D disparity and noise removal problem. This results in homogeneous appearances within anatomical structures which is exploited further in a Random Forest framework. Context-aware appearance models are learnt using Random Forest (RF) for appearance-based prediction of great vessel and blood pool of an unseen subject during testing. We have obtained promising results on the HVSMR16 training dataset in a leave-one-out cross-validation. © Springer International Publishing AG 2017.
Noack M.,Zuse Institute Berlin
ACM International Conference Proceeding Series | Year: 2017
For writing a new scientific application, portability across existing and future hardware should be the major design goal, as there is a multitude of different compute devices, and programme codes typically outlive systems by far. Unlike other programming models that address parallelism or heterogeneity, OpenCL does provide practical portability across a wide range of HPC-relevant architectures. Other than that, it has a range of further advantages like being a library-only implementation, and using runtime kernelcompilation. We present experiences with utilising OpenCL alongside C++, MPI, and CMake in two real-world scienti€c codes. Our targets are a Cray XC40 supercomputer with multi-And many-core (Xeon Phi) CPUs, as well as multiple smaller systems with NVIDIA and AMD GPUs. We shed light on practical issues arising in such a scenario, like the interaction between OpenCL and MPI, discuss solutions, and point out current limitations of OpenCL in the domain of scientific HPC from an application developer's and user's point of view. © 2017 ACM.
Costa M.,University of Cambridge |
Manton J.D.,University of Cambridge |
Ostrovsky A.D.,University of Cambridge |
Ostrovsky A.D.,University of Heidelberg |
And 3 more authors.
Neuron | Year: 2016
Neural circuit mapping is generating datasets of tens of thousands of labeled neurons. New computational tools are needed to search and organize these data. We present NBLAST, a sensitive and rapid algorithm, for measuring pairwise neuronal similarity. NBLAST considers both position and local geometry, decomposing neurons into short segments; matched segments are scored using a probabilistic scoring matrix defined by statistics of matches and non-matches. We validated NBLAST on a published dataset of 16,129 single Drosophila neurons. NBLAST can distinguish neuronal types down to the finest level (single identified neurons) without a priori information. Cluster analysis of extensively studied neuronal classes identified new types and unreported topographical features. Fully automated clustering organized the validation dataset into 1,052 clusters, many of which map onto previously described neuronal types. NBLAST supports additional query types, including searching neurons against transgene expression patterns. Finally, we show that NBLAST is effective with data from other invertebrates and zebrafish. Video Abstract © 2016 MRC Laboratory of Molecular Biology
Pomplun J.,Zuse Institute Berlin |
Schmidt F.,Zuse Institute Berlin
SIAM Journal on Scientific Computing | Year: 2010
We propose a new method for fast estimation of error bounds for outputs of interest in the reduced basis context, efficiently applicable to real world 3D problems. Geometric parameterizations of complicated 2D, or even simple 3D, structures easily leads to affine expansions consisting of a high number of terms (oc 100-1000). Applicat ion of state-of-the-art techniques for computation of error bounds becomes practically impossible. As a way out we propose a new error estimator, inspired by the subdomain residuum method, which leads to substantial savings (orders of magnitude) regarding online and offline computational times and memory consumption. We apply certified reduced basis techniques with the newly developed error estimator to 3D electromagnetic scattering problems on unbounded domains. A numerical example from computational lithography demonstrates the good performance and effectivity of the proposed estimator. © 2010 Society for Industrial and Applied Mathematics.
Weiser M.,Zuse Institute Berlin
BIT Numerical Mathematics | Year: 2014
Spectral deferred correction methods for solving stiff ODEs are known to converge reasonably fast towards the collocation limit solution on equidistant grids, but show a less favourable contraction on non-equidistant grids such as Radau-IIa points. We interprete SDC methods as fixed point iterations for the collocation system and propose new DIRK-type sweeps for stiff problems based on purely linear algebraic considerations. Good convergence is recovered also on non-equidistant grids. The properties of different variants are explored on a couple of numerical examples. © 2014 Springer Science+Business Media Dordrecht
Schiela A.,Zuse Institute Berlin
Mathematical Programming | Year: 2013
We propose and analyze an interior point path-following method in function space for state constrained optimal control. Our emphasis is on proving convergence in function space and on constructing a practical path-following algorithm. In particular, the introduction of a pointwise damping step leads to a very efficient method, as verified by numerical experiments. © 2012 Springer-Verlag Berlin Heidelberg and Mathematical Optimization Society.
Weber M.,Zuse Institute Berlin
Studies in Classification, Data Analysis, and Knowledge Organization | Year: 2013
In this chapter, PCCAC is described as a special spectral clustering algorithm which is applicable for molecular simulation data. From a mathematical point of view, only PCCAC is able to correctly identify the physical timescales of molecular motion. In order to decrease the statistical error of this timescales analysis, an adaptive clustering algorithm is necessary. © Springer-Verlag Berlin Heidelberg 2013.
Duy N.T.,Zuse Institute Berlin
Medical image computing and computer-assisted intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention | Year: 2012
We propose a fully automatic method for tooth detection and classification in CT or cone-beam CT image data. First we compute an accurate segmentation of the maxilla bone. Based on this segmentation, our method computes a complete and optimal separation of the row of teeth into 16 subregions and classifies the resulting regions as existing or missing teeth. This serves as a prerequisite for further individual tooth segmentation. We show the robustness of our approach by providing extensive validation on 43 clinical head CT scans.
Berthold T.,Zuse Institute Berlin
Operations Research Letters | Year: 2013
In modern MIP solvers, primal heuristics play a key role in finding high-quality solutions. However, classical performance measures reflect the impact of primal heuristics on the overall solving process badly. In this article, we introduce a new performance measure, the "primal integral", which depends on the quality of solutions and on the time when they are found. We compare five state-of-the-art MIP solvers w.r.t. the newly proposed measure, and show that heuristics improve their performance by up to 80%.© 2013 Elsevier B.V.
Berthold T.,Zuse Institute Berlin |
Gleixner A.M.,Zuse Institute Berlin
Mathematical Programming | Year: 2014
We present Undercover, a primal heuristic for nonconvex mixed-integer nonlinear programs (MINLPs) that explores a mixed-integer linear subproblem (sub-MIP) of a given MINLP. We solve a vertex covering problem to identify a smallest set of variables to fix, a so-called cover, such that each constraint is linearized. Subsequently, these variables are fixed to values obtained from a reference point, e.g., an optimal solution of a linear relaxation. Each feasible solution of the sub-MIP corresponds to a feasible solution of the original problem. We apply domain propagation to try to avoid infeasibilities, and conflict analysis to learn additional constraints from infeasibilities that are nonetheless encountered. We present computational results on a test set of mixed-integer quadratically constrained programs (MIQCPs) and MINLPs. It turns out that the majority of these instances allows for small covers. Although general in nature, we show that the heuristic is most successful on MIQCPs. It nicely complements existing root-node heuristics in different state-of-the-Art solvers and helps to significantly improve the overall performance of the MINLP solver SCIP. © 2013 Springer-Verlag Berlin Heidelberg and Mathematical Optimization Society.