Search

When it comes to enhancing exploitation of massive data, machine learning and AI methods are very much at the forefront of our awareness. Much less so is the need for, and complexity of, applying these techniques efficiently across memory-distributed data volumes. Heat [1, 2] is an open-source Python library for high- performance data analytics, machine learning, and deep learning. It provides highly optimized algorithms and data structures for tensor computations using CPUs, GPUs and distributed cluster systems. Heat's Numpy-like API makes writing scalable, GPU-accelerated applications straightforward - at the same time, parallelism implemented under the hood via MPI provides a significant improvement in efficiency and performance with respect to, e.g., Dask. Born out of a large-scale collaboration in applied sciences, Heat also acts a platform for collaboration and knowledge transfer within data-intensive science. In this presentation, I will show you the inner workings of the library, tell you about our collaborations with the astrophysics and space science community (massively parallel signal-processing capabilities for the SKA- MPG telescope among others) and hopefully gain from you some insight into how to best support data- intensive astro operations going forward. References: [1] Gotz, M., Debus, C., Coquelin, et al.: "HeAT - a Distributed and GPU-accelerated Tensor Framework for Data Analytics"; [2] https://github.com/helmholtz-analytics/heat

Born out of a large-scale collaboration in applied sciences, Heat [1, 2] is an open-source Python library for high-performance data analytics, machine learning, and deep learning. It provides highly optimized algorithms and data structures for tensor computations using CPUs, GPUs and distributed cluster systems. Plus, with Heat, writing scalable scientific and data science applications is as straightforward as using NumPy. With as diverse a user base as, e.g., the Earth System Modeling, neuroscience, and aerospace research communities, Heat offers not only generalized solutions for data-intensive science, but also a platform for ever-expanding cross-discipline collaborations and knowledge transfer. We look forward to many interactions and possible collaborations at AI STAR. [1] Götz, M., Debus, C., Coquelin, et al.: "HeAT - a Distributed and GPU-accelerated Tensor Framework for Data Analytics". 2020 IEEE International Conference on Big Data (Big Data) (pp. 276-287) [2] https://github.com/helmholtz-analytics/heat &nbsp

HeAT is a distributed tensor framework for high performance data analytics. The goal of HeAT is to fill the gap between machine learning libraries that have a strong focus on exploiting GPUs for performance, and traditional, distributed high-performance computing (HPC). The basic idea is to provide a generic, distributed tensor library with machine learning methods based on it.

We present the Helmholtz Analytics Toolkit (HeAT), a scientific big data analytics library for HPC systems. The large progress in big data analytics in general and machine learning/deep learning (ML/DL) in particular, has been considerably enforced by well-designed open source libraries like Hadoop, Spark, Storm, Disco, scikit-learn, H2O.ai, Mahout, TensorFlow, PaddlePaddle, PyTorch, Caffe, Keras, MXNet, CNTK, BigDL, Theano, Neon, Chainer, DyNet, Dask and Intel DAAL to mention only a few of them. Despite the large number of existing data analytics frameworks, a library taking the specific needs in scientific big data analytics under consideration is still missing. For instance, no pre-existing library operates on heterogeneous hardware like GPU/CPU systems while allowing transparent computation on distributed systems. Typical big data analytics frameworks like Spark are designed for distributed memory systems and consequently do not fully explore the shared memory architecture as well as the network technology of HPC systems. ML/DL frameworks like Theano or Chainer focus on single node computations or when providing mechanisms for distributed computation, as done by TensorFlow or PytTorch, they impose the details of the distributed computation to the programmer. Libraries designed for HPC like Dask and Intel DAAL do not provide any GPU support. The presented library - HeAT - is designed for the specific needs of big data analytics in the scientific context. It is based on a tensor data object on which operations can be performed like basic scalar functions, linear algebra algorithms, slicing or broadcasting operations necessary for most data analytics algorithms. The tensor data objects reside either on the CPU or on the GPU and, if needed, is distributed over different nodes. Operations on tensor objects are transparent to the user, i.e. they keep the same irrespective whether the tensor object resides on a single node or if it is distributed over several nodes allowing conveniently port algorithms from single nodes to multiple nodes or from CPUs to GPUs. HeAT builds in top of PyTorch which already provides many required features like automatic differentiation, CPU and GPU support, linear algebra operations and basic MPI functionality as well as an imperative programming paradigm allowing for fast prototyping essentially in scientific research. In addition to basic tensor operations, HeAT implements several common data analytics algorithms, e.g. kmeans, logistic regression and neural networks, optimized for large scale distributed systems. After motivating the framework and specifying its scope, we will describe its concept and its realization in detail. We demonstrate its usage by means of several typical examples from data analytics including benchmarks comparing HeAT to pre-existing frameworks. The presentation closes with a discussion on the downsides, further developments and future challenges of HeAT.

Dissertation, RWTH Aachen University, 2018; Aachen 1 Online-Ressource (xiii,123 Seiten) : Illustrationen, Diagramme (2018). doi:10.18154/RWTH-2018-229719 = Dissertation, RWTH Aachen University, 2018, Free-surface flows are ubiquitous in our lives. Almost all flows we encounter as human beings are of this type. However, when we think of flows we expect mostly water flows. The latter is an example of a Newtonian flow, which only show viscous behavior. Subject of this thesis, however, are the viscoelastic flows, which - in addition to the viscous behavior - behave elastically to a certain degree. In many cases, this elasticity is due to a polymeric component in the fluid. This thesis will take the reader from the microscopic world of polymer chains to the macroscopic description of a polymer solution or polymer melt. Moreover, the thesis highlights important properties that carry over from the microscopic to the macroscopic world. One of these properties is that the conformation tensor should stay positive definite. The latter then subsequently motivates the derivation of a numerical scheme, based on the work of Fattal and Kupferman, that by replacing the conformation tensor with the log-conformation tensor in the numerical model ensures this positivity by design. The novelty of our approach is that it allows for the easier application of many analytical tools. E.g., Newton's method can be easily applied to the non-linear equation system, but also the numerical analysis of schemes becomes easier. The central application, in which this thesis culminates, is the simulation of a rising air bubble in a viscoelastic fluid. The latter, so simple it may seem, shows phenomena that are to date not fully understood. It has long been observed experimentally that the bubble velocity is not always a continuous function of the bubble volume. This thesis tries to shed some new light on the mechanism behind this discontinuity by showing the emergence of vortices on the bubble surface., Published by Aachen

Several studies imply that anodal transcranial direct current stimulation (tDCS) over the left dorsolateral prefrontal cortex (DLPFC) can modulate the formation of verbal episodic memories. The aim of this study was to test if tDCS through a multi-electrode Laplacian montage over the left DLPFC could differentially modulate declarative memory performance depending on the application phase. Two groups of healthy participants (n = 2 × 15) received 1 mA anodal or sham stimulation for 20 min during the encoding or during the recall phase on a delayed cued-recall, using a randomized, double-blinded, repeated-measures experimental design. Memory performance was assessed at two time points: 10 min and 24 h after learning. We found no significant difference between anodal and sham stimulation with regard to the memory scores between conditions (stimulation during encoding or recall) or between time points, suggesting that anodal tDCS over the left DLPFC with these stimulation parameters had no effect on the encoding and the consolidation of associative verbal content.

Masterarbeit, RWTH Aachen University, 2015; Aachen 71 S. (2015). = Masterarbeit, RWTH Aachen University, 2015, Published by Aachen