Research

Projects

• DFG project “Multi-fidelity, Active Learning Strategies for Exciton Transfer Among Adsorbed Molecules” (duration 2022-2025)
  Research project within the DFG SPP 2363 “Utilization and Development of Machine Learning for Molecular Applications – Molecular Machine Learning” (joint project with Ulrich Kleinekathöfer, Physics, Jacobs University Bremen)
• MarDATA project “Bayesian Chronology Modelling for Paleoclimate Archives” (duration: 2022-2025)
  Helmholtz School for Marine Data Science (joint project with Florian Adolphi, Alfred Wegener Institute).
• MarDATA project “Digital Ice Cores: Paleo-Climate reconstruction using Bayesian methods” (duration: 2022-2025)
  Helmholtz School for Marine Data Science (joint project with Thomas Laepple, Alfred Wegener Institute & U. Bremen).
• DFG project “Excitation Energy Transfer in a Photosynthetic System with more than 100 Million Atoms” (duration: 2021-2024)
  Individual Research Grant funded by DFG (joint project with Ulrich Kleinekathöfer, Physics, Jacobs University Bremen)

Research interests

• Machine Learning, Uncertainty Quantification and Big Data
  • multi-fidelity machine learning (e.g. by sparse grid combination technique)
  • approximate training (low-rank approximation by e.g. hierarchical matrices)
  • stochastic collocation, Bayesian inference / data assimilation
  • basic research wrt. reproducing kernel Hilbert spaces / Gaussian processes
• High Performance Computing
  • numerics / algorithms for many-core processors (e.g. GPUs)
  • scalable distributed-memory parallel computing in machine learning and scientific computing
• Interdisciplinary applications
  • material science, quantum chemistry (data from DFT, CC, etc.)
  • paleo-climate reconstruction (calibration, etc.)
  • fluid mechanics (two-phase flows, plasma physics)
  • medical imaging (dynamic contrast-enhanced imaging)

Software

• hmglib – Open Source library for hierarchical matrices on GPU (https://github.com/zaspel/hmglib)
• MPLA – Open Source multi-GPU parallel library for dense iterative solvers (https://github.com/zaspel/MPLA)
• QML – Extension of the Open Source Quantum Machine Learning library by multi-fidelity techniques (for now hosted in the fork https://github.com/zaspel/qml)
• UQ on GPU – (Multi-) GPU parallel kernel-based methods for uncertainty quantification in random partial differential equations
• AMG on GPU – Hybrid GPU-based Ruge-Stüben algebraic multigrid
• NaSt3DGPF – Multi-GPU MPI-parallel version of the fluid mechanics solver NaSt3DGPF (https://ins.uni-bonn.de/media/public/u/griebel/NaSt3DGPF/projects.html)

Showcases