Synergies

Within the DSSC several institutes at the University of Groningen collaborate on topics such as:

Complex system modelling and reconstruction

At the DSSC the Johann Bernoulli Institute for Mathematics and Computer Science (JBI) and the Engineering and Technology Institute Groningen will work on time-series reconstruction and model identification, extreme events modelling, and large-scale statistical inference for high-dimensional data, with applications to (low-order) climate models, turbulence, and metabolic regulation.

The Kapteyn Astronomical Institute and JBI will collaborate on models of formation and evolution of galaxies and the cosmic web, black hole formation and growth, and star formation.

New collaborations will be initiated with the Groningen Research Institute of Pharmacy (GRIP) in the area of causal inference modelling in epidemiology.Learning of models is a central topic of many institutes, with many possible interactions.

JBI and the Institute for Artificial Intelligence and Cognitive Engineering work on machine learning, learning vector quantization, continuous learning, reinforcement learning; adaptivity and cognition (e.g., Monk system for handwriting recognition), and meta-learning (increasing learning effectiveness); statistical analysis and validation of multivariate data; visual analytics of machine learning algorithms; and (non)linear dimension reduction. Kapteyn and JBI collaborate on data clustering of data (e.g., GAIA database), extracting information from multidimensional data (e.g., APERTIF-Westerbork), and 3D map of the Milky Way (GAIA space telescope). An important class of models concerns networks: important topics are dynamical modelling, evolution and statistical inference of networks; (stochastic) dynamics on complex networks; adaptive network protocols; multi-agent simulations and emergent behaviour (collaboration with theoretical biology).

JBI and Kapteyn collaborate on geometrical learning, learning of cosmological models, and structure detection in the large scale universe (cosmic web) by geometric and topological methods. New collaboration will be initiated with GRIP on machine learning techniques for multi-omics clinical data.

Large-scale computing and visualization

Within this topic, the JBI works on computational methods for fluid flows and multiphysics systems, and simulation of partial differential equation models; efficient approaches for large scale statistical inference; tera-pixel image processing; processing of large medical data and network visualization (with UMCG-NIC); and brain-inspired computing (EU Human Brain Project).

Under the auspices of the DSSC researchers from JBI further collaborate with Kapteyn researchers on feature extraction and visual exploration of high-dimensional data sets, and e-visualization of Big Data. Kapteyn works on LOFAR Epoch of Reionization Project; N-body simulations; gravitational lensing; and adaptive mesh refinement simulations.

The Van Swinderen Institute for Particle Physics and Gravity works on and with computational methods for high-energy physics (Monte Carlo, large sparse matrices, statistical mechanics, statistics).

JBI and ALICE collaborate on cognitive, perceptual, and user-adaptive visualization. Kapteyn and JBI work on visualization of galaxy evolution (example: GAIA project); morphological analysis of galactic structures and the cosmic web; visualization of cosmic structure (formation); query-driven visualization, and interactive visual exploration of astronomical catalogues.

Kapteyn, ALICE, and JBI collaborate, partly through the TARGET project, on the processing of huge image collections, and the processing and visualization of medical scans for disease prediction (GLIMPS). New ways of interacting with large data are explored, for example by making use of touch displays and virtual 3d environments.

The Centre for Information Technology (CIT) offers visualisation solutions at various levels of collaboration, i.e.: (i) at the researcher’s desk (individual); (ii) in the Reality Centre at the CIT (in teams); and (iii) in Infoversum’s Big Data Dome (large group interventions).

JBI and GRIP plan new work on causal and predictive visualization of high-dimensional prescription databases, and visualization of genomics, proteomics and metabolomics networks. The Zernike Institute for Advanced Materials (ZIAM), Stratingh, and the Groningen Biomolecular Sciences and Biotechnology Institute work together in the FOM Focus group for developing new materials for organic photovoltaics.

Future particle accelerator facilities such as FAIR require novel feature extraction techniques for the online reduction of massively streaming data coming from a complex network of various types of radiation sensors operating autonomously at high speed. The KVI–CART has taken up this challenge by developing efficient pattern recognition and feature extraction algorithms that will be deployed on parallel computing architectures, and that will process and correlate the information of each sensor with the aim to provide a significant data reduction. Subsequently, the pre-processed and filtered data are subject to ‘needle in a haystack’ searches running on a distributed computing environment with the aim to identify the events of interest, reduce the dimensionality of the data, and to extract observables with a high precision and accuracy. The development of geometry-based particle-track-finding algorithms is performed in collaboration with JBI. In contrast to the more conventionally known image-processing scenarios, the data generated from these accelerator-driven instruments are a massive collection of small-sized (tens of kilobytes) images (‘events’) adding up to tens of petabytes of data each year and produced at high frequencies. Each event is characterized by its high dimensionality due to the complexity of the underlying physics process and the variety of sensors. The data are made accessible for large international research communities such as BESIII, NUSTAR, and PANDA.

New treatment modalities in radiation oncology, such as proton radiotherapy, aim at a further reduction of the radiation-induced complications, which still are frequently a factor limiting treatment outcome. The calculation of both the treatment plan itself and the information needed for verification of proper delivery of the treatment are using Monte Carlo methods and require the analysis of multi-modality (MRI, X-ray CT, PET/SPECT) 3D imaging data to determine the properties of the tissues traversed by the radiation. Because of the sensitivity of the quality of treatment delivery on small changes in the patient anatomy, frequent reoptimization of the treatment plan during the 4-6 week treatment course is required. The KVI-Centre for Advanced Radiation Technology (KVI-CART) medical physics group works on the development of efficient methods for both the Monte Carlo calculations and the analysis of the imaging data, which are key to the success of these new treatment modalities. With the ever increasing computational capacity the full simulation of complex accelerator systems comes within reach. These simulations incorporate all the external electromagnetic fields for guiding and accelerating the particles, space charge effects, wakefields generated by the particles, synchrotron radiation, halo formation due to particle matter interaction and manufacturing and installation tolerances. More efficient algorithms are an important step towards multi-bunch calculations, thus giving insight in bunch-bunch interactions, with realistic numbers of particles per bunch. These calculations are particularly important for CW free electron lasers, on which the KVI-CART accelerator physics group will start working in collaboration with industry in the near future.