A5035572633- Scientific Computing and Data Management
- Parallel Computing and Optimization Techniques
- Advanced Data Storage Technologies
- Distributed and Parallel Computing Systems
- Oceanographic and Atmospheric Processes
- Hydrology and Watershed Management Studies
- Geophysics and Gravity Measurements
- Meteorological Phenomena and Simulations
- Research Data Management Practices
- Environmental Monitoring and Data Management
- Advanced Neural Network Applications
- Climate variability and models
- Advanced Image and Video Retrieval Techniques
- Hydrological Forecasting Using AI
- Cloud Computing and Resource Management
- Radio Astronomy Observations and Technology
- Advanced Electron Microscopy Techniques and Applications
- Machine Learning and Data Classification
- Advanced Multi-Objective Optimization Algorithms
- Advanced Fluorescence Microscopy Techniques
- Computational Physics and Python Applications
- Stochastic Gradient Optimization Techniques
- scientometrics and bibliometrics research
- Metaheuristic Optimization Algorithms Research
- Digital Media Forensic Detection
Netherlands eScience Center
2015-2024
Leiden University
2024
Centrum Wiskunde & Informatica
2022
Vrije Universiteit Amsterdam
2010-2014
Many GPU applications perform data transfers to and from memory at regular intervals. For example because the does not fit into or of internode communication end each time step. Overlapping computation with CPU-GPU can reduce costs moving data. Several different techniques exist for transferring overlapping those computation. It is currently known when apply which method. Implementing benchmarking method often a large programming effort feasible. To solve these issues provide insight in...
A very common problem in GPU programming is that some combination of thread block dimensions and other code optimization parameters, like tiling or unrolling factors, results dramatically better performance than kernel configurations. To obtain highly-efficient kernels it often required to search vast discontinuous spaces consist all possible combinations values for tunable parameters. This paper presents Kernel Tuner, an easy-to-use tool testing auto-tuning OpenCL, CUDA, C with support many...
Abstract Single molecule localization microscopy offers in principle resolution down to the molecular level, but practice this is limited primarily by incomplete fluorescent labeling of structure. This missing information can be completed merging from many structurally identical particles. In work, we present an approach for 3D single particle analysis which hugely increases signal-to-noise ratio and enables determining symmetry groups macromolecular complexes. Our method does not require a...
Researchers are often faced with exploring new research domains. Broad questions about the domain, such as who influential authors or what important topics, difficult to answer due overwhelming number of relevant publications. Therefore, we present litstudy: a Python package that enables answering using simple scripts Jupyter notebooks. The selecting scientific publications and studying their metadata visualizations, bibliographic network analysis, natural language processing. software was...
Abstract. Hutton et al. (2016) argued that computational hydrology can only be a proper science if the hydrological community makes sure model studies are executed and presented in reproducible manner. Hut, Drost van de Giesen replied to achieve this hydrologists should not “re-invent water wheel” but rather use existing technology from other fields (such as containers ESMValTool) open interfaces Basic Model Interface, BMI) do their (Hut al., 2017). With paper associated release of...
Finding optimal parameter configurations for tunable GPU kernels is a non-trivial exercise large search spaces, even when automated. This poses an optimization task on nonconvex space, using expensive to evaluate function with unknown derivative. These characteristics make good candidate Bayesian Optimization, which has not been applied this problem before. However, the application of Optimization challenging. We demonstrate how deal rough, discrete, constrained containing invalid...
Abstract. As an extreme scenario of dynamical sea level changes, regional surface height (SSH) changes that occur in the North Atlantic due to abrupt weakening meridional overturning circulation (AMOC) are simulated. Two versions same ocean-only model used study effect ocean resolution on these SSH changes: a high-resolution (HR) strongly eddying version and low-resolution (LR) which eddies is parameterised. The AMOC induced both by applying strong freshwater perturbations around Greenland....
Abstract. The Parallel Ocean Program (POP) is used in many strongly eddying ocean circulation simulations. Ideally it would be desirable to able do thousand-year-long simulations, but the current performance of POP prohibits these types In this work, using a new distributed computing approach, two methods improve are presented. first block-partitioning scheme for optimization load balancing such that can run efficiently multi-platform setting. second implementation part model code on...
Recent years have witnessed phenomenal growth in the application, and capabilities of Graphical Processing Units (GPUs) due to their high parallel computation power at relatively low cost. However, writing a computationally efficient GPU program (kernel) is challenging, generally only certain specific kernel configurations lead significant increases performance. Auto-tuning process automatically optimizing software for highly-efficient execution on target hardware platform. particularly...
Abstract. In this paper we present the Oceanographic Multipurpose Software Environment (OMUSE). OMUSE aims to provide a homogeneous environment for existing or newly developed numerical ocean simulation codes, simplifying their use and deployment. way, experiments that combine models representing different physics spanning ranges of physical scales can be easily designed. Rapid development is made possible through creation simple high-level scripts. The low-level core abstraction in designed...
Detecting single-pulse astronomical transients, such as pulsars or fast radio bursts, requires collecting and processing enormous amounts of data. AMBER is a software pipeline that makes it possible to detect these phenomena in real-time. The achieves this by offloading compute-intensive kernels many-core accelerators. Additionally, automatically tunes achieve high performance on variety different platforms. We therefore see an important tool the search for new interesting transients.
We present a series of optimizations to alleviate stack memory overflow issues and improve overall performance GPU computational kernels in atmospheric chemical kinetics model simulations. use heap numerical solvers for stiff ODEs, move reaction constants tracer concentration arrays from global memory, direct pointer indexing array access, CUDA streams overlap computation with transfer the device. Overall, an order magnitude reduction requirements is achieved, allowing simultaneous...
Adapting applications to optimally utilize available hardware is no mean feat: the plethora of choices for optimization techniques are infeasible maximize manually. To this end, auto-tuning frameworks used automate task, which in turn use algorithms efficiently search vast searchspaces. However, there a lack comparability studies presenting advances and incorporated. As each publication varies way experiments conducted, metrics used, results reported, comparing performance among publications...
Analyzing digital images is an important investigation in forensics with the ever increasing number of from computers and smartphones. In this article we aim to advance state-of-the-art common image source identification (which originate same camera). To end, present two types applications for different goals that make use a) a modern Desktop computer GPU b) highly heterogeneous cluster many kinds GPUs, something call computing jungles. The first application targets medium-scale...
3D digital city models, important for urban planning, are currently constructed from massive point clouds obtained through airborne LiDAR (Light Detection and Ranging). They semantically enriched with information auxiliary GIS data like Cadastral which contains about the boundaries of properties, road networks, rivers, lakes etc.