PUMA publications for /user/testusersimtech/miscMon Jan 27 13:09:00 CET 2025ChamLarge-Scale Scientific Computations99--107Randomized Symplectic Model Order Reduction for Hamiltonian Systems2024EXC2075 PN3 PN3A-8 PN5 PN5-7 PN6 PN6-1(II) misc Simulations of large scale dynamical systems in multi-query or real-time contexts require efficient surrogate modelling techniques, as e.g. achieved via Model Order Reduction (MOR). Recently, symplectic methods like the complex singular value decomposition (cSVD) or the SVD-like decomposition have been developed for preserving Hamiltonian structure during MOR. In this contribution, we show how symplectic structure preserving basis generation can be made more efficient with randomized matrix factorizations. We present a randomized complex SVD (rcSVD) algorithm and a randomized SVD-like decomposition (rSVD-like). We demonstrate the efficiency of the approaches with numerical experiments on high dimensional systems.Mon Jan 27 13:09:00 CET 2025Explaining Vision-Language Similarities in Dual Encoders with Feature-Pair Attributions2024EXC2075 PN6-5 PN6-5(II) misc Mon Jan 27 13:09:00 CET 2025Learned RESESOP for solving inverse problems with inexact forward operator2024EXC2075 PN1 PN1-10 misc Mon Jan 27 13:09:00 CET 20252022 IEEE Spoken Language Technology Workshop (SLT)912--919Anonymizing Speech with Generative Adversarial Networks to Preserve Speaker Privacy2023EXC2075 PN6 PN6-5 PN6-5(II) misc Mon Jan 27 13:09:00 CET 2025PreprintA Deep Learning Approach for Thermal Plume Prediction of Groundwater Heat Pumps2022EXC2075 PN6 PN6-2 misc Mon Jan 27 13:09:00 CET 2025PreprintDeep learning based surrogate modeling for thermal plume prediction of groundwater heat pumps2023EXC2075 PN6 PN6-2 misc Mon Jan 27 13:09:00 CET 2025Error Analysis of Randomized Symplectic Model Order Reduction for Hamiltonian systems2024EXC2075 PN3 PN3A-8 PN5 PN5-7 PN6 PN6-1(II) misc Mon Jan 27 13:09:00 CET 2025New York, NY, USAProceedings of the 12th International Workshop on OpenCL and SYCL041-4IWOCL '24Evaluation of SYCL’s Different Data Parallel Kernels2024EXC2075 PN6 PN6-2(II) misc SYCL provides programmers with four, and in the case of AdaptiveCpp even five, ways for calling and writing a device kernel. This paper analyzes the performance of these diverse kernel invocation types for DPC++ and AdaptiveCpp as SYCL implementations on an NVIDIA A100 GPU, an AMD Instinct MI210 GPU, and a dual-socket AMD EPYC 9274F CPU. Using the example of a kernel matrix assembly, we show why the performance can differ by a factor of 100 in the worst case on the same hardware for the same problem using different SYCL implementations and kernel invocation types.Mon Jan 27 13:09:00 CET 2025INTERSPEECH 20234788--4792Controllable Generation of Artificial Speaker Embeddings through Discovery of Principal Directions2023EXC2075 PN6-5 PN6-5(II) misc Mon Oct 07 09:24:07 CEST 20242023 32nd International Conference on Computer Communications and Networks (ICCCN)07Persival: Using Delayed Remote Updates in a Distributed Mobile Simulation2023EXC2075 PN7-1(II)Becker PN7-1.1 misc Executing complex simulations on mobile devices such as augmented reality (AR) glasses or smartphones enables many novel pervasive applications. For instance, a physiotherapist can display muscles and bones in real-time as a visual overlay on the patient's body. The major challenge of such pervasive simulations is the complexity of the simulation, which typically exceeds the resources of the mobile device by far. Offloading computationally intensive simulations to a remote server is a promising method to enable real-time simulations on resource-constrained mobile devices without compromising the quality of the simulation results. However, the results of offloaded computations may arrive with an inevitable communication delay, which is critical for real-time simulations and also induces communication overhead. In this work, we tackle these challenges by proposing a novel approach for pervasive simulations on mobile devices. We combine a low-quality local Neural Network (NN) model on the mobile device with a high-quality NN model on a remote server, particularly taking care to integrate delayed updates from the server with the local simulation results. This distributed approach has several advantages over purely local or remote execution models: We benefit from high-quality remote results, while being robust to dynamic delays, server and network failures, and we reduce the communication overhead.Mon Oct 07 09:24:07 CEST 2024Development of efficient multiscale multiphysics models accounting for reversible flow at various subsurface energy storage sites2021EXC2075 PN5 PN5-2A misc Mon Sep 16 14:37:42 CEST 2024High Performance Computing in Science and Engineering{'} 17: Transactions of the High Performance Computing Center, Stuttgart (HLRS) 2017241-254Numerical analysis of heat transfer during cooling of supercritical fluid by means of direct numerical simulation2018EXC2075 PN1 PN1-7 misc Mon Sep 16 14:37:42 CEST 2024High Performance Computing in Science and Engineering '22An Investigation of Information Flux between Turbulent Boundary Layer and Porous Medium2024EXC2075 PN1 PN1-7 misc Mon Sep 16 14:37:42 CEST 2024International Journal of Heat and Mass Transfer06122626Celebration of Professor Bernhard Weigand on his 60th birthday1882022EXC2075 PN1 PN1-7 misc Mon Sep 16 14:37:42 CEST 2024Quantum computing through the lens of control: A tutorial introduction2023EXC2075 PN8 misc Thu Sep 12 12:02:31 CEST 2024keine PublikationTraining robust and generalizable quantum models2024EXC2075 PN8 misc Adversarial robustness and generalization are both crucial properties of reliable machine learning models. In this paper, we study these properties in the context of quantum machine learning based on Lipschitz bounds. We derive parameter-dependent Lipschitz bounds for quantum models with trainable encoding, showing that the norm of the data encoding has a crucial impact on the robustness against data perturbations. Further, we derive a bound on the generalization error which explicitly involves the parameters of the data encoding. Our theoretical findings give rise to a practical strategy for training robust and generalizable quantum models by regularizing the Lipschitz bound in the cost. Further, we show that, for fixed and non-trainable encodings, as those frequently employed in quantum machine learning, the Lipschitz bound cannot be influenced by tuning the parameters. Thus, trainable encodings are crucial for systematically adapting robustness and generalization during training. The practical implications of our theoretical findings are illustrated with numerical results. Thu Sep 12 12:02:31 CEST 202405Projekttag fehltModelling Turbulence in Coupled Environments: The K-Omega Shear Stress Transport ModelMaster's Thesis2021misc Thu Sep 12 12:02:31 CEST 2024Hydrology and Earth System Scienceskeine PublikationTowards a community-wide effort for benchmarking in subsurface hydrological inversion: benchmarking cases, high-fidelity reference solutions, procedure and a first comparison2024 (submitted)EXC2075 PN5 misc Thu Sep 12 12:02:31 CEST 202409Projekttag fehltCoupled Turbulent Free- and Porous Media Flows: Investigations of Interfacial RoughnessMaster's Thesis2022misc Thu Sep 12 12:02:31 CEST 2024arXivnur PreprintData-driven identification of latent port-Hamiltonian systems2024EXC2075 PN3 PN3A-8 PN7 PN7-6(II) misc